Pith. sign in

REVIEW 4 major objections 5 minor 171 references

This paper claims that real-world MCP server faults cluster into five empirically derived categories, and a practitioner survey confirms all five occur in practice.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

T0 review · deepseek-v4-flash

2026-08-04 05:51 UTC pith:DGD3Z35D

load-bearing objection Useful first taxonomy of MCP server faults, but the 'comprehensive' framing outruns the evidence; still worth a serious referee. the 4 major comments →

arxiv 2603.05637 v2 pith:DGD3Z35D submitted 2026-03-05 cs.SE

Real Faults in Model Context Protocol (MCP) Software: a Comprehensive Taxonomy

classification cs.SE
keywords Model Context ProtocolMCP serverssoftware fault taxonomyLLM-based systemsempirical software engineeringissue miningtool integrationconfiguration faults
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This study tries to establish that the faults that actually break Model Context Protocol (MCP) servers are not random or idiosyncratic: they cluster into five recognizable categories. Analyzing 407 bug-related closed issues from 385 open-source MCP server projects, the authors derive a taxonomy whose largest categories are server/tool configuration, server/host configuration, and server setting, followed by documentation and general programming. A survey of 41 MCP practitioners reports encountering every category in practice, which the authors take as evidence the taxonomy is complete. If right, the taxonomy gives developers a checklist of where MCP systems fail and gives researchers a baseline for MCP-specific testing and repair tools.

Core claim

The central claim is that the first empirically grounded taxonomy of MCP server faults consists of five high-level categories: Server Setting (27.45% of labeled issues), Server/Tool Configuration (31.74%), Server/Host Configuration (28.64%), Documentation (6.92%), and General Programming (5.25%). The study also finds that MCP-specific faults differ measurably from non-MCP faults: they attract more comments and more comments per collaborator, yet are fixed faster and by developers with less overall experience. The authors interpret the survey results as confirming that all taxonomy categories occur in practice, with tool response handling and documentation reported most frequently and tool di

What carries the argument

The load-bearing object is the taxonomy itself, anchored in MCP's host-client-server architecture. Faults are classified according to where in the MCP stack the break happens: in the server's own setting, in how the server wires tools, in how the host connects to the server, in documentation, or in ordinary programming errors. The taxonomy works by taking issue reports and assigning each to a category based on the component boundary where the fault manifests, which is what lets the study compare fault types and separate MCP-specific problems from generic ones.

Load-bearing premise

The claim rests on the assumption that the 407 issues selected by the LLM-assisted classification and clustering pipeline faithfully represent real MCP faults; the pipeline's accuracy was checked on only 40 hand-labeled issues, and clustering-relatedness was judged on 20 issues per cluster, so sampling or labeling errors could reshape the taxonomy.

What would settle it

Label a fresh random sample of 200 closed bug issues from MCP server projects written in non-Python SDKs using the paper's codebook. If a substantial fraction (say more than 15-20%) of MCP-related bugs fall outside the five categories, or if the category distribution shifts dramatically, the taxonomy's completeness and generalizability claims fail.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • MCP server testing efforts should target configuration and integration seams—dependencies, tool registration, host connection settings—rather than only tool logic.
  • Issue triage tools for MCP repositories can use the five categories as a checklist to route and prioritize bug reports.
  • MCP-specific fault patterns being distinct from generic bugs means automated debugging and repair should be trained on MCP-aware data, not generic software faults.
  • The observed short fix times and lower fixer experience for MCP faults suggest a small pool of domain specialists handles them; improving documentation and defaults could reduce that bottleneck.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The taxonomy's three dominant categories are all configuration seams, suggesting that MCP's fragility lives at integration boundaries rather than in the core protocol logic; a testable consequence is that configuration-focused static analysis would catch most avoidable failures.
  • The survey's finding that tool discovery/registration is rare but rated critical implies that availability failures outweigh frequency when prioritizing MCP reliability; one could monitor issue trackers to test whether discovery faults cause outsized user impact.
  • Because the sample is Python-only and collected soon after MCP's release, the taxonomy may underrepresent faults that emerge with maturity, such as security issues and version skew across SDKs; extending the same coding to TypeScript SDKs and later time windows is a natural check.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper presents an empirical study of faults in Model Context Protocol (MCP) servers. The authors mine 385 GitHub repositories that use the Python MCP SDK, extract 26,821 closed issues, classify them with an LLM (GPT-4o-mini) into bug and non-bug categories, cluster the 3,282 bug issues with BERTopic, and manually identify 407 MCP-related issues. These issues are open-coded to produce a five-category taxonomy: Server Setting, Server/Tool Configuration, Server/Host Configuration, Documentation, and General Programming, with a detailed subcategory structure. The taxonomy is said to be validated by a survey of 41 MCP practitioners, and the paper further compares MCP-related versus non-MCP faults on metrics such as fix time, number of comments, collaborators, and fixer experience. The central claim is that this is the first large-scale, comprehensive taxonomy of real faults in MCP servers, and that all identified categories occur in practice.

Significance. If the taxonomy is sound, this is a valuable contribution to the emerging area of MCP reliability engineering. The paper has clear strengths: the methodology is transparently described, the taxonomy is data-derived rather than imposed a priori, concrete issue examples ground each category, a replication package is promised, and the survey explicitly involves practitioners who did not construct the taxonomy. The statistical analyses report effect sizes, not just p-values, which is good practice. However, the degree to which the taxonomy is 'comprehensive' depends on the recall of the automated filtering pipeline, which is not measured. The small survey and its recruitment from the same repository population limit the strength of the validation. These concerns do not invalidate the taxonomy as a useful initial structure, but they do affect the strength of the central claim and the generalizability of the quantitative proportions.

major comments (4)
  1. [§3.2.2 and §6.2] The MCP-relatedness filter has no measured recall. Clusters are labeled MCP-related by manually reviewing 20 representative issues per cluster, and issues in non-MCP clusters are recovered only if their LLM-generated summary contains the literal token 'MCP'. Because issue titles and summaries often omit the acronym (e.g., 'server crashes on startup' in an MCP-server repository), MCP-specific issues assigned to non-MCP clusters are silently dropped. This directly affects the 407-issue sample, the taxonomy structure, and the category proportions in Fig. 3. The paper's threat-to-validity discussion acknowledges this but does not quantify the risk. Please add a recall estimate (e.g., manually inspect a random sample from non-MCP clusters or from issues whose summaries do not contain 'MCP') or explicitly soften the 'comprehensive' claim to one about the population of issues that are explicitl
  2. [§4.1 and §4.3 (Tables 5 and 6)] The statistical comparison of fault characteristics treats labels as independent units. The paper reports 419 labels from 383 issues after excluding 24 issues, stating that some issues received multiple labels. For metrics such as 'required time to fix,' an issue with two labels contributes to both groups in the Kruskal–Wallis and Dunn tests, violating the independence assumption and potentially inflating significance or distorting effect sizes. Please use a primary label per issue for the RQ3 analysis, or otherwise account for the multi-label structure. This is important because the significant differences reported in Table 6 may not be robust.
  3. [§3.2.3] The open-coding step is the sole basis for the taxonomy, but no inter-rater reliability statistic is reported. The paper states this is because no predefined labels were available. While this is understandable for emergent coding, the central claim would be substantially strengthened by reporting the number of disagreements resolved, a post-hoc agreement check on a random subset using the final codebook, or a more detailed justification of why a reliability measure is impossible. Without this, readers cannot distinguish robust categories from the systematic bias of the two raters. The current description of regular meetings and third-adjudicator resolution is helpful, but it does not quantify reliability.
  4. [§3.3 and §4.2 (Table 4)] The survey is presented as validating the completeness of the taxonomy, but the evidence is weak. There are 41 valid responses, recruited from contributors to the same repositories that produced the issue sample; the response rate is not reported. Table 4 shows that every category is encountered by at least some respondents, but this only confirms that each category is non-empty; it does not establish completeness or external generalizability. The absence of newly suggested fault types from 41 respondents is not strong evidence of coverage. Please reframe the survey as a preliminary qualitative check, report recruitment and response-rate information, and avoid the conclusion that it 'supports the completeness' of the taxonomy.
minor comments (5)
  1. [§7 (Conclusion)] The conclusion says '443 MCP server repositories,' while §3.1 reports 385 repositories after manual checking. This inconsistency should be fixed.
  2. [§4.1] The relationship between 407 inspected issues, 24 excluded issues, and 419 labels should be stated more clearly. The current text says 'Each issue was examined and labeled... resulting in 419 labels' and then mentions 24 exclusions; readers may wonder whether the percentages are computed on labels or issues.
  3. [Figure 2] The methodology figure contains the typo 'MPC-based Repos'; it should read 'MCP-based Repos'.
  4. [Table 4] The table reports prevalence percentages without giving the denominator for each row. Since the number of respondents who answered each category group may differ, include N per row or clarify how missing responses are handled.
  5. [§3.2.2] The paper says HDBSCAN was used to estimate the number of clusters and then KMeans was applied with k=101. It is not clear how the HDBSCAN result maps to 101; a brief description or a sensitivity check would improve reproducibility.

Circularity Check

0 steps flagged

No significant circularity: the taxonomy is derived from manual labeling of issue reports and validated by independent survey respondents.

full rationale

The paper's central deliverable is an empirically grounded taxonomy of faults in MCP servers. The derivation chain is: collect repositories using the MCP Python SDK; filter them; classify closed issues with an LLM selected on 40 hand-labeled examples; cluster bug summaries with BERTopic; manually review 20 issues per cluster plus keyword fallback to identify 407 MCP-related issues; open-code those issues into a codebook; group labels into a hierarchy; and validate via a survey of 41 MCP practitioners. None of these steps defines a taxonomy category in terms of the survey outcome, nor does the survey define the categories: survey respondents were not the taxonomy constructors, were asked whether they had encountered each pre-existing fault type, and were invited to suggest new types. The manual labeling and codebook construction in Section 3.2.3 are independent of the survey validation. Self-citations such as [114], [116], and [117] are methodological (exclusion criteria, metric definitions, taxonomy-construction procedure) rather than load-bearing theoretical premises, and no uniqueness theorem or prior result by the same authors is invoked to force the taxonomy's structure. The main limitations—unmeasured recall in the LLM/clustering filter (acknowledged in Section 6.2 as a risk of false negatives) and recruitment of survey participants from the same repository ecosystem—affect external validity and confidence in completeness, but they are not cases where a 'prediction' is equivalent to an input by construction. Accordingly, no circular step is identified.

Axiom & Free-Parameter Ledger

6 free parameters · 6 axioms · 0 invented entities

No mathematical free parameters or invented entities are introduced. The listed free parameters are data-selection and methodological thresholds that shape the corpus, the 407-issue sample, and the survey validation. The axioms are the assumptions needed to treat mined GitHub issues, LLM-assisted labeling, clustering, manual coding, and a small survey as evidence for a comprehensive taxonomy.

free parameters (6)
  • min_stars_and_forks = 10
    Repositories with fewer than 10 stars and 10 forks were excluded (§3.1), shaping the corpus of MCP servers from which all faults were drawn.
  • date_threshold = 2024-11-01
    Only closed issues after Nov 1, 2024 were kept (§3.1), because MCP was officially introduced then; this truncates the fault history.
  • number_of_clusters_k = 101
    KMeans clustering into 101 clusters guided which issues were considered MCP-related; the taxonomy depends on this grouping (§3.2.2).
  • reviewed_issues_per_cluster = 20
    Raters manually reviewed 20 representative issues per cluster to decide MCP-relatedness; this selection threshold determines the 407-issue sample (§3.2.2).
  • llm_ground_truth_size = 40
    GPT-4o-mini was chosen for issue classification based on 40 manually labeled issues; classifier accuracy is assumed to generalize (§3.2.1, Table 1).
  • survey_sample_size = 41 valid responses
    Completeness validation rests on 41 self-selected respondents from 10,756 invited email addresses (§4.2).
axioms (6)
  • domain assumption Closed GitHub issues, after LLM triage, are a valid proxy for real MCP server faults.
    The entire taxonomy is mined from closed issues labeled as bugs; issue text may misreport root cause or be environment-specific (§3.2.1, §6.1).
  • domain assumption GPT-4o-mini's issue-classification performance on 40 ground-truth issues (F1≈0.77) generalizes to 26,821 issues.
    No second validation set is reported; misclassification propagates into the bug set and the taxonomy (§3.2.1, Table 1).
  • domain assumption BERTopic clustering accurately separates MCP-related from non-MCP issues so the 407-issue sample is representative.
    Cluster membership and 20-issue manual review determine MCP-relatedness; false negatives are acknowledged in §6.2.
  • domain assumption Open coding by two raters without computing inter-rater agreement yields stable categories.
    Section 3.2.3 states agreement cannot be calculated because no predefined labels; this weakens the reliability of the final taxonomy categories.
  • domain assumption 41 survey respondents are representative of MCP practitioners.
    Self-selected respondents, mostly developers, drawn from the same repository pool; used to claim completeness (§4.2).
  • standard math Non-parametric statistical tests (Kruskal–Wallis, Mann–Whitney, Dunn) with reported effect sizes are appropriate for the metric distributions.
    Used per §4.3 after Shapiro–Wilk rejected normality; standard statistical tools.

pith-pipeline@v1.3.0-alltime-deepseek · 37628 in / 13285 out tokens · 134327 ms · 2026-08-04T05:51:35.661044+00:00 · methodology

0 comments
read the original abstract

The rapid adoption of foundation models has significantly expanded the capabilities of software systems, enabling them to perform complex language, reasoning, and interaction tasks that were previously difficult to automate. However, this progress has also introduced novel challenges that were largely absent in previous generations of software. In particular, the increasing integration of foundation models with external tools and resources raises new concerns regarding reliability, security, and robustness. The Model Context Protocol (MCP) has recently been proposed to standardize interactions between AI-based software systems, software tools, and external resources. Despite its growing adoption, there remains limited systematic understanding of real-world faults in MCP-based software systems. In this paper, we present the first large-scale taxonomy of faults in MCP servers, comprising five high-level fault categories derived from empirical evidence. To evaluate the completeness and generalizability of this taxonomy, we conduct a survey of MCP practitioners with diverse roles and experience levels. The results confirm that all identified fault categories occur in practice and reveal distinct characteristics that differentiate MCP-specific faults from non-MCP faults. Overall, this study provides actionable insights for researchers and practitioners by identifying the most error-prone and critical components of MCP-based systems. These insights can inform the development of more robust, reliable, and secure AI-enabled software systems that rely on MCP.

Figures

Figures reproduced from arXiv: 2603.05637 by Foutse Khomh, Mina Taraghi, Mohammad Mehdi Morovati.

Figure 1
Figure 1. Figure 1: A high-level view of Model Context Protocol (MCP) Architecture [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: High-level view of the used methodology files within defined ranges (using the parameter ‘size:<min>..<max>’) to ensure that each query returned fewer than 1000 results. Since the search API do not return any results for file sizes greater than 350 KB in our case, we divide the entire range from 1 B to 350 KB into intervals of 50 B. This process result in a total of approximately 7,000 API calls. We then e… view at source ↗
Figure 3
Figure 3. Figure 3: Taxonomy of MCP server issues (counts in parentheses). [PITH_FULL_IMAGE:figures/full_fig_p012_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Comparison of faults’ characteristics for the 5 main categories of bugs in terms of metrics including [PITH_FULL_IMAGE:figures/full_fig_p025_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Comparison of faults’ characteristics for MCP vs. Non-MCP [PITH_FULL_IMAGE:figures/full_fig_p027_5.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

171 extracted references · 1 canonical work pages

  1. [1]

    Data Mining: Concepts and Techniques (Third Edition)

    2012. Data Mining: Concepts and Techniques (Third Edition). Morgan Kaufmann, Boston, xxiii–xxix. doi:10.1016/B978- 0-12-381479-1.00020-4

  2. [2]

    ISO/IEC/IEEE International Standard - Systems and software engineering–Vocabulary.ISO/IEC/IEEE 24765:2017(E)(2017), 1–541

    2017. ISO/IEC/IEEE International Standard - Systems and software engineering–Vocabulary.ISO/IEC/IEEE 24765:2017(E)(2017), 1–541. doi:10.1109/IEEESTD.2017.8016712

  3. [3]

    image.no_webcam_support·Issue #10143·gradio-app/gradio

    2024. image.no_webcam_support·Issue #10143·gradio-app/gradio. https://github.com/gradio-app/gradio/issues/ 10143. [Accessed 16-12-2025]

  4. [4]

    MCP tool doesn’t show up in Claude (macOS)·Issue #12·MarkusPfundstein/mcp-gsuite

    2024. MCP tool doesn’t show up in Claude (macOS)·Issue #12·MarkusPfundstein/mcp-gsuite. https://github.com/ MarkusPfundstein/mcp-gsuite/issues/12

  5. [5]

    Architecture overview

    2025. Architecture overview. https://modelcontextprotocol.io/docs/learn/architecture. [Accessed 16-12-2025]

  6. [6]

    (Bedrock Knowledge Base Retrieval MCP): (403 Error when accessing Bedrock Knowledge Base through QueryKnowledgeBases tool)·Issue #237·awslabs/mcp

    2025. (Bedrock Knowledge Base Retrieval MCP): (403 Error when accessing Bedrock Knowledge Base through QueryKnowledgeBases tool)·Issue #237·awslabs/mcp. https://github.com/awslabs/mcp/issues/237

  7. [7]

    [BUG] Dot in Project Path Causes Errors·Issue #226·basicmachines-co/basic-memory

    2025. [BUG] Dot in Project Path Causes Errors·Issue #226·basicmachines-co/basic-memory. https://github.com/ basicmachines-co/basic-memory/issues/226

  8. [8]

    [BUG] Latest MCP docker image fails due to missing Azure-identity package·Issue #778·getzep/graphiti

    2025. [BUG] Latest MCP docker image fails due to missing Azure-identity package·Issue #778·getzep/graphiti. https://github.com/getzep/graphiti/issues/778

  9. [9]

    [BUG]: Unity Catalog Server goes directly to ’crashed’ state in Databricks Apps·Issue #28·databrickslabs/mcp

    2025. [BUG]: Unity Catalog Server goes directly to ’crashed’ state in Databricks Apps·Issue #28·databrickslabs/mcp. https://github.com/databrickslabs/mcp/issues/28

  10. [10]

    callback http servers should not run forever·Issue #4·pipeboard-co/meta-ads-mcp

    2025. callback http servers should not run forever·Issue #4·pipeboard-co/meta-ads-mcp. https://github.com/ pipeboard-co/meta-ads-mcp/issues/4

  11. [11]

    Cannot install dicom-mcp and dicom-mcp[dev]==0.1.1 because these package versions have conflicting dependencies·Issue #14·ChristianHinge/dicom-mcp

    2025. Cannot install dicom-mcp and dicom-mcp[dev]==0.1.1 because these package versions have conflicting dependencies·Issue #14·ChristianHinge/dicom-mcp. https://github.com/ChristianHinge/dicom-mcp/issues/14

  12. [12]

    Cannot run mcp tools with google-adk agents·Issue #13·Rudra-ravi/wikipedia-mcp

    2025. Cannot run mcp tools with google-adk agents·Issue #13·Rudra-ravi/wikipedia-mcp. https://github.com/Rudra- ravi/wikipedia-mcp/issues/13

  13. [13]

    Claude Code hook configuration broken - multi-agent file locking non-functional·Issue #32·rinadelph/Agent- MCP

    2025. Claude Code hook configuration broken - multi-agent file locking non-functional·Issue #32·rinadelph/Agent- MCP. https://github.com/rinadelph/Agent-MCP/issues/32

  14. [14]

    Claude completely lost it: all schemas fail·Issue #71·rusiaaman/wcgw

    2025. Claude completely lost it: all schemas fail·Issue #71·rusiaaman/wcgw. https://github.com/rusiaaman/wcgw/ issues/71

  15. [15]

    Claude Desktop can’t connect·Issue #10·Rudra-ravi/wikipedia-mcp

    2025. Claude Desktop can’t connect·Issue #10·Rudra-ravi/wikipedia-mcp. https://github.com/Rudra-ravi/wikipedia- mcp/issues/10

  16. [16]

    connection refuse·Issue #621·PDFMathTranslate/PDFMathTranslate

    2025. connection refuse·Issue #621·PDFMathTranslate/PDFMathTranslate. https://github.com/PDFMathTranslate/ PDFMathTranslate/issues/621. [Accessed 16-12-2025]

  17. [17]

    Duplicate version tool in MCP tool list·Issue #120·BeehiveInnovations/zen-mcp-server

    2025. Duplicate version tool in MCP tool list·Issue #120·BeehiveInnovations/zen-mcp-server. https://github.com/ BeehiveInnovations/zen-mcp-server/issues/120

  18. [18]

    entity_id gets "correct" to wrong string·Issue #11·voska/hass-mcp

    2025. entity_id gets "correct" to wrong string·Issue #11·voska/hass-mcp. https://github.com/voska/hass-mcp/ issues/11 J. ACM, Vol. 37, No. 4, Article 111. Publication date: August 2026. Real Faults in Model Context Protocol (MCP) Software: a Comprehensive Taxonomy 111:33

  19. [19]

    ‘ERROR: Exception in ASGI application‘ when running MCP with Docker·Issue #470·getzep/graphiti

    2025. ‘ERROR: Exception in ASGI application‘ when running MCP with Docker·Issue #470·getzep/graphiti. https://github.com/getzep/graphiti/issues/470

  20. [20]

    Fatal exception: Project name ’XXXXX’ already exists and points to <some other repository>·Issue #314· oraios/serena

    2025. Fatal exception: Project name ’XXXXX’ already exists and points to <some other repository>·Issue #314· oraios/serena. https://github.com/oraios/serena/issues/314

  21. [21]

    get_conversation and list_conversations tools not exposed despite being defined in code·Issue #51· elevenlabs/elevenlabs-mcp

    2025. get_conversation and list_conversations tools not exposed despite being defined in code·Issue #51· elevenlabs/elevenlabs-mcp. https://github.com/elevenlabs/elevenlabs-mcp/issues/51

  22. [22]

    Gmail Batch API Causes Consistent SSL/TLS Crashes in Python MCP Server·Issue #115·taylorwilsdon/google_- workspace_mcp

    2025. Gmail Batch API Causes Consistent SSL/TLS Crashes in Python MCP Server·Issue #115·taylorwilsdon/google_- workspace_mcp. https://github.com/taylorwilsdon/google_workspace_mcp/issues/115

  23. [23]

    Incorrect Redis password environment variable name in the sample Claude Desktop configuration·Issue #6· redis/mcp-redis

    2025. Incorrect Redis password environment variable name in the sample Claude Desktop configuration·Issue #6· redis/mcp-redis. https://github.com/redis/mcp-redis/issues/6

  24. [24]

    Invalid JSON parsing when starting on Windows environment·Issue #12·doobidoo/mcp-memory-service

    2025. Invalid JSON parsing when starting on Windows environment·Issue #12·doobidoo/mcp-memory-service. https://github.com/doobidoo/mcp-memory-service/issues/12

  25. [25]

    MCP Server returns fact_embedding arrays causing 50x token inflation·Issue #610·getzep/graphiti

    2025. MCP Server returns fact_embedding arrays causing 50x token inflation·Issue #610·getzep/graphiti. https: //github.com/getzep/graphiti/issues/610

  26. [26]

    Multiple sessions: ‘The review seems to be looking at a different repository again‘·Issue #28· BeehiveInnovations/zen-mcp-server

    2025. Multiple sessions: ‘The review seems to be looking at a different repository again‘·Issue #28· BeehiveInnovations/zen-mcp-server. https://github.com/BeehiveInnovations/zen-mcp-server/issues/28

  27. [27]

    [OpenMemory] Cannot connect, 404 on ‘messages‘ route, wrong URL (?)·Issue #2695·mem0ai/mem0

    2025. [OpenMemory] Cannot connect, 404 on ‘messages‘ route, wrong URL (?)·Issue #2695·mem0ai/mem0. https://github.com/mem0ai/mem0/issues/2695

  28. [28]

    2025. Prompts. https://modelcontextprotocol.io/specification/2025-11-25/server/prompts. [Accessed 16-12-2025]

  29. [29]

    REDDIT_USERNAME and REDDIT_PASSWORD are not optional for readonly operations·Issue #4· Arindam200/reddit-mcp

    2025. REDDIT_USERNAME and REDDIT_PASSWORD are not optional for readonly operations·Issue #4· Arindam200/reddit-mcp. https://github.com/Arindam200/reddit-mcp/issues/4

  30. [30]

    Resources

    2025. Resources. https://modelcontextprotocol.io/specification/2025-11-25/server/resources. [Accessed 16-12-2025]

  31. [31]

    2025. SDKs. https://modelcontextprotocol.io/docs/sdk. [Accessed 16-12-2025]

  32. [32]

    Server crashes on startup due to f-string syntax error·Issue #15·elevenlabs/elevenlabs-mcp

    2025. Server crashes on startup due to f-string syntax error·Issue #15·elevenlabs/elevenlabs-mcp. https://github. com/elevenlabs/elevenlabs-mcp/issues/15

  33. [33]

    Should not do prints in console when using STDIO transport·Issue #4·Rudra-ravi/wikipedia-mcp

    2025. Should not do prints in console when using STDIO transport·Issue #4·Rudra-ravi/wikipedia-mcp. https: //github.com/Rudra-ravi/wikipedia-mcp/issues/4

  34. [34]

    Too slow when save from Claude with openmemory·Issue #3011·mem0ai/mem0

    2025. Too slow when save from Claude with openmemory·Issue #3011·mem0ai/mem0. https://github.com/mem0ai/ mem0/issues/3011

  35. [35]

    2025. Tools. https://modelcontextprotocol.io/specification/2025-11-25/server/tools. [Accessed 16-12-2025]

  36. [36]

    Unable to connect to local mcp server over sse, when using MCP Inspector·Issue #18·elevenlabs/elevenlabs-mcp

    2025. Unable to connect to local mcp server over sse, when using MCP Inspector·Issue #18·elevenlabs/elevenlabs-mcp. https://github.com/elevenlabs/elevenlabs-mcp/issues/18

  37. [37]

    Unable to use any tool due to parameter error·Issue #19·echelon-ai-labs/servicenow-mcp

    2025. Unable to use any tool due to parameter error·Issue #19·echelon-ai-labs/servicenow-mcp. https://github. com/echelon-ai-labs/servicenow-mcp/issues/19

  38. [38]

    Understanding MCP clients

    2025. Understanding MCP clients. https://modelcontextprotocol.io/docs/learn/client-concepts. [Accessed 16-12-2025]

  39. [39]

    Understanding MCP servers

    2025. Understanding MCP servers. https://modelcontextprotocol.io/docs/learn/server-concepts. [Accessed 16-12- 2025]

  40. [40]

    with claude-code the find_symbol and related mcp tools don’t return any results·Issue #153·oraios/serena

    2025. with claude-code the find_symbol and related mcp tools don’t return any results·Issue #153·oraios/serena. https://github.com/oraios/serena/issues/153

  41. [41]

    2015.Data mining: the textbook

    Charu C Aggarwal et al. 2015.Data mining: the textbook. Vol. 1. Springer

  42. [42]

    Emad Aghajani, Csaba Nagy, Mario Linares-Vásquez, Laura Moreno, Gabriele Bavota, Michele Lanza, and David C Shepherd. 2020. Software documentation: the practitioners’ perspective. InProceedings of the acm/ieee 42nd interna- tional conference on software engineering. 590–601

  43. [43]

    Roozbeh Aghili, Heng Li, and Foutse Khomh. 2023. Studying the characteristics of AIOps projects on GitHub. Empirical Software Engineering28, 6 (2023), 143

  44. [44]

    Khairul Alam and Banani Roy. 2026. Analyzing GitHub Issues and Pull Requests in nf-core Pipelines: Insights into nf-core Pipeline Repositories.arXiv preprint arXiv:2601.09612(2026)

  45. [45]

    Khairul Alam, Banani Roy, Chanchal K Roy, and Kartik Mittal. 2025. An empirical investigation on the challenges in scientific workflow systems development.Empirical Software Engineering30, 5 (2025), 151

  46. [46]

    Mohammad Aldhaen. 2020. Interview versus questionnaire from the perspective of CBE members.International Journal of Education, Learning and Development8, 2 (2020), 21–41

  47. [47]

    2024.Introducing the Model Context Protocol

    Anthropic. 2024.Introducing the Model Context Protocol. https://www.anthropic.com/news/model-context-protocol

  48. [48]

    Armstrong

    Richard A. Armstrong. 2014. When to use the Bonferroni correction.Ophthalmic and Physiological Optics34, 5 (2014), 502–508. doi:10.1111/opo.12131

  49. [49]

    Azure-Samples. 2025. Azure-Samples/semantic-kernel-advanced-usage. https://github.com/Azure-Samples/semantic- kernel-advanced-usage J. ACM, Vol. 37, No. 4, Article 111. Publication date: August 2026. 111:34 Taraghi et al

  50. [50]

    Amiangshu Bosu and Jeffrey C Carver. 2014. Impact of developer reputation on code review outcomes in oss projects: An empirical investigation. InProceedings of the 8th ACM/IEEE international symposium on empirical software engineering and measurement. 1–10

  51. [51]

    Morakot Choetkiertikul, Peerachai Banyongrakkul, Chaiyong Ragkhitwetsagul, Suppawong Tuarob, Hoa Khanh Dam, and Thanwadee Sunetnanta. 2024. Sprint2Vec: a deep characterization of sprints in iterative software development. IEEE Transactions on Software Engineering51, 1 (2024), 220–242

  52. [52]

    2013.Statistical Power Analysis for the Behavioral Sciences(2 ed.)

    Jacob Cohen. 2013.Statistical Power Analysis for the Behavioral Sciences(2 ed.). Routledge, New York. doi:10.4324/ 9780203771587

  53. [53]

    GitHub developer guideline documentation. 2021. GitHub REST API. https://developer.github.com/v3/. Accessed: 2025-09-05

  54. [54]

    Yihong Dong, Xue Jiang, Jiaru Qian, Tian Wang, Kechi Zhang, Zhi Jin, and Ge Li. 2025. A survey on code generation with llm-based agents.arXiv preprint arXiv:2508.00083(2025)

  55. [55]

    Olive Jean Dunn. 1964. Multiple Comparisons Using Rank Sums.Technometrics6, 3 (Aug. 1964), 241–252. doi:10.1080/00401706.1964.10490181 Publisher: ASA Website _eprint: https://www.tandfonline.com/doi/pdf/10.1080/00401706.1964.10490181

  56. [56]

    Abul Ehtesham, Aditi Singh, Gaurav Kumar Gupta, and Saket Kumar. 2025. A survey of agent interoperability protocols: Model Context Protocol (MCP), Agent Communication Protocol (ACP), Agent-to-Agent Protocol (A2A), and Agent Network Protocol (ANP). doi:10.48550/arXiv.2505.02279 arXiv:2505.02279 [cs]

  57. [57]

    Kenneth Enevoldsen, Isaac Chung, Imene Kerboua, Márton Kardos, Ashwin Mathur, David Stap, Jay Gala, Wissam Siblini, Dominik Krzemiński, Genta Indra Winata, Saba Sturua, Saiteja Utpala, Mathieu Ciancone, Marion Schaeffer, Diganta Misra, Shreeya Dhakal, Jonathan Rystrøm, Roman Solomatin, Ömer Veysel Çağatan, Akash Kundu, Martin Bernstorff, Shitao Xiao, Aksh...

  58. [58]

    Rosa Falotico and Piero Quatto. 2015. Fleiss’ kappa statistic without paradoxes.Quality & Quantity49, 2 (2015), 463–470

  59. [59]

    Daocheng Fu, Xin Li, Licheng Wen, Min Dou, Pinlong Cai, Botian Shi, and Yu Qiao. 2024. Drive like a human: Rethinking autonomous driving with large language models. In2024 IEEE/CVF Winter Conference on Applications of Computer Vision Workshops (W ACVW). IEEE, 910–919

  60. [60]

    2004.Software quality assurance: from theory to implementation

    Daniel Galin. 2004.Software quality assurance: from theory to implementation. Pearson education

  61. [61]

    Ahmad Nauman Ghazi, Kai Petersen, Sri Sai Vijay Raj Reddy, and Harini Nekkanti. 2018. Survey research in software engineering: Problems and mitigation strategies.IEEE Access7 (2018), 24703–24718

  62. [62]

    Google. [n. d.]. Embeddings | Gemini API. https://ai.google.dev/gemini-api/docs/embeddings

  63. [63]

    Maarten Grootendorst. 2022. BERTopic: Neural topic modeling with a class-based TF-IDF procedure. doi:10.48550/ arXiv.2203.05794 arXiv:2203.05794 [cs]

  64. [64]

    Maarten Grootendorst. 2025. MaartenGr/BERTopic. https://github.com/MaartenGr/BERTopic original-date: 2020-09- 22T14:19:29Z

  65. [65]

    JSON-RPC Working Group. 2010. JSON-RPC 2.0 Specification. https://www.jsonrpc.org/specification. [Accessed 16-12-2025]

  66. [66]

    Hechuan Guo, Yongle Hao, Yue Zhang, Minghui Xu, Peizhuo Lv, Jiezhi Chen, and Xiuzhen Cheng. 2025. A Measurement Study of Model Context Protocol Ecosystem. doi:10.48550/arXiv.2509.25292 arXiv:2509.25292 [cs]

  67. [67]

    Zikang Guo, Benfeng Xu, Chiwei Zhu, Wentao Hong, Xiaorui Wang, and Zhendong Mao. 2025. MCP-AgentBench: Evaluating Real-World Language Agent Performance with MCP-Mediated Tools. doi:10.48550/arXiv.2509.09734 arXiv:2509.09734 [cs]

  68. [68]

    Duc Minh Ha, Phu Trac Kien, Tho Quan, and Anh Nguyen-Duc. 2025. Evaluating Classical Software Process Models as Coordination Mechanisms for LLM-Based Software Generation.arXiv preprint arXiv:2509.13942(2025)

  69. [69]

    Muhammad Usman Hadi, Rizwan Qureshi, Abbas Shah, Muhammad Irfan, Anas Zafar, Muhammad Bilal Shaikh, Naveed Akhtar, Jia Wu, Seyedali Mirjalili, et al . 2023. Large language models: a comprehensive survey of its applications, challenges, limitations, and future prospects.Authorea preprints1, 3 (2023), 1–26. J. ACM, Vol. 37, No. 4, Article 111. Publication d...

  70. [70]

    Mohammed Mehedi Hasan, Hao Li, Emad Fallahzadeh, Gopi Krishnan Rajbahadur, Bram Adams, and Ahmed E Hassan. 2025. Model context protocol (mcp) at first glance: Studying the security and maintainability of mcp servers. arXiv preprint arXiv:2506.13538(2025)

  71. [71]

    HazyResearch. 2026. HazyResearch/minions. https://github.com/HazyResearch/minions

  72. [72]

    Kim Herzig, Sascha Just, and Andreas Zeller. 2013. It’s not a bug, it’s a feature: how misclassification impacts bug prediction. In2013 35th international conference on software engineering (ICSE). IEEE, 392–401

  73. [73]

    Maria Cristina Hinojosa Lee, Johan Braet, and Johan Springael. 2024. Performance metrics for multilabel emotion classification: comparing micro, macro, and weighted f1-scores.Applied Sciences14, 21 (2024), 9863

  74. [74]

    Xinyi Hou, Yanjie Zhao, Shenao Wang, and Haoyu Wang. 2025. Model context protocol (mcp): Landscape, security threats, and future research directions.arXiv preprint arXiv:2503.23278(2025)

  75. [75]

    Xinyi Hou, Yanjie Zhao, Shenao Wang, and Haoyu Wang. 2025. Model Context Protocol (MCP): Landscape, Security Threats, and Future Research Directions. doi:10.48550/arXiv.2503.23278 arXiv:2503.23278 [cs]

  76. [76]

    Hugging Face. 2024. gemma-7b. https://huggingface.co/google/gemma-7b. Accessed: 10/09/2025

  77. [77]

    Hugging Face. 2024. Llama-3.1-8B. https://huggingface.co/meta-llama/Llama-3.1-8B. Accessed: 10/09/2025

  78. [78]

    Hugging Face. 2024. zephyr-7b. https://huggingface.co/HuggingFaceH4/zephyr-7b-beta. Accessed: 10/09/2025

  79. [79]

    Hugging Face. 2025. Mistral-7B-v0.1. https://huggingface.co/mistralai/Mistral-7B-v0.1. Accessed: 10/09/2025

  80. [80]

    Hugging Face. 2025. Qwen3-8B. https://huggingface.co/Qwen/Qwen3-8B. Accessed: 10/09/2025

Showing first 80 references.