mcp-servers (Six production-minded MCP servers)
Provides repository intelligence through public GitHub APIs, including read-only access to repository info, language, topics, README, CI, commits, issues, and code-search.
One typed model across public greenhouse job boards as part of the ats-jobs integration, enabling interaction with ATS job posting APIs.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@mcp-servers (Six production-minded MCP servers) Evaluate this candidate's CV against the Senior ML Engineer job posting"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Six production-minded MCP servers
60-second quickstart
git clone https://github.com/prathamesh-git9/mcp-servers.git
cd mcp-servers
uv sync --all-packages --dev
uv run grounded-cv-eval
uv run pytest -qThat runs a real BM25 + dense + reciprocal-rank-fusion evaluation and the entire
socket-restricted test suite. Each server is directly runnable with uv run grounded-cv,
uv run repo-intel, uv run web-research, uv run ats-jobs,
uv run outcome-ledger, or uv run coding-workflows.
Related MCP server: MCP Agent System
claude_desktop_config.json
Copy this complete block. uvx installs the two required workspace packages from
the public repository and caches the environment locally.
{
"mcpServers": {
"grounded-cv": {
"command": "uvx",
"args": [
"--from",
"git+https://github.com/prathamesh-git9/mcp-servers.git#subdirectory=packages/grounded-cv",
"--with",
"git+https://github.com/prathamesh-git9/mcp-servers.git#subdirectory=packages/common",
"grounded-cv"
]
},
"repo-intel": {
"command": "uvx",
"args": [
"--from",
"git+https://github.com/prathamesh-git9/mcp-servers.git#subdirectory=packages/repo-intel",
"--with",
"git+https://github.com/prathamesh-git9/mcp-servers.git#subdirectory=packages/common",
"repo-intel"
]
},
"web-research": {
"command": "uvx",
"args": [
"--from",
"git+https://github.com/prathamesh-git9/mcp-servers.git#subdirectory=packages/web-research",
"--with",
"git+https://github.com/prathamesh-git9/mcp-servers.git#subdirectory=packages/common",
"web-research"
]
},
"ats-jobs": {
"command": "uvx",
"args": [
"--from",
"git+https://github.com/prathamesh-git9/mcp-servers.git#subdirectory=packages/ats-jobs",
"--with",
"git+https://github.com/prathamesh-git9/mcp-servers.git#subdirectory=packages/common",
"ats-jobs"
]
},
"outcome-ledger": {
"command": "uvx",
"args": [
"--from",
"git+https://github.com/prathamesh-git9/mcp-servers.git#subdirectory=packages/outcome-ledger",
"--with",
"git+https://github.com/prathamesh-git9/mcp-servers.git#subdirectory=packages/common",
"outcome-ledger"
]
},
"coding-workflows": {
"command": "uvx",
"args": [
"--from",
"git+https://github.com/prathamesh-git9/mcp-servers.git#subdirectory=packages/coding-workflows",
"--with",
"git+https://github.com/prathamesh-git9/mcp-servers.git#subdirectory=packages/common",
"coding-workflows"
]
}
}
}.mcp.json
{
"mcpServers": {
"grounded-cv": {
"type": "stdio",
"command": "uvx",
"args": [
"--from",
"git+https://github.com/prathamesh-git9/mcp-servers.git#subdirectory=packages/grounded-cv",
"--with",
"git+https://github.com/prathamesh-git9/mcp-servers.git#subdirectory=packages/common",
"grounded-cv"
]
},
"repo-intel": {
"type": "stdio",
"command": "uvx",
"args": [
"--from",
"git+https://github.com/prathamesh-git9/mcp-servers.git#subdirectory=packages/repo-intel",
"--with",
"git+https://github.com/prathamesh-git9/mcp-servers.git#subdirectory=packages/common",
"repo-intel"
]
},
"web-research": {
"type": "stdio",
"command": "uvx",
"args": [
"--from",
"git+https://github.com/prathamesh-git9/mcp-servers.git#subdirectory=packages/web-research",
"--with",
"git+https://github.com/prathamesh-git9/mcp-servers.git#subdirectory=packages/common",
"web-research"
]
},
"ats-jobs": {
"type": "stdio",
"command": "uvx",
"args": [
"--from",
"git+https://github.com/prathamesh-git9/mcp-servers.git#subdirectory=packages/ats-jobs",
"--with",
"git+https://github.com/prathamesh-git9/mcp-servers.git#subdirectory=packages/common",
"ats-jobs"
]
},
"outcome-ledger": {
"type": "stdio",
"command": "uvx",
"args": [
"--from",
"git+https://github.com/prathamesh-git9/mcp-servers.git#subdirectory=packages/outcome-ledger",
"--with",
"git+https://github.com/prathamesh-git9/mcp-servers.git#subdirectory=packages/common",
"outcome-ledger"
]
},
"coding-workflows": {
"type": "stdio",
"command": "uvx",
"args": [
"--from",
"git+https://github.com/prathamesh-git9/mcp-servers.git#subdirectory=packages/coding-workflows",
"--with",
"git+https://github.com/prathamesh-git9/mcp-servers.git#subdirectory=packages/common",
"coding-workflows"
]
}
}
}Architecture
flowchart LR
C[Claude Desktop / MCP client] -->|JSON-RPC over stdio| G[grounded-cv]
C --> R[repo-intel]
C --> W[web-research]
C --> A[ats-jobs]
C --> O[outcome-ledger]
C --> CW[coding-workflows]
G --> H[BM25 + 384-D dense vectors + RRF]
G --> P[(Committed structured profile)]
R --> GH[Public GitHub REST API]
W --> RB[URL policy + robots.txt + extractor]
A --> ATS[Six public ATS APIs]
O --> SQ[(SQLite WAL / FULL sync)]
CW --> RC[Local repo structure + configured quality gates]
G & R & W & A & O & CW --> CORE["Shared typed failures, deadlines, redaction,<br/>rate limiting and cache"]The repository is a seven-package workspace: one small shared core and one independently installable package for each server. Every server exposes tools, resources, and at least one prompt over the official MCP Python SDK's stdio transport.
What a hiring manager can verify quickly
Server | One-line proof of work |
grounded-cv | Structured profile resources, genuine hybrid RAG, exact-span citations, and conservative claim verification. |
repo-intel | Read-only repository, language, topic, README, CI, commit, issue, and code-search intelligence through public GitHub APIs. |
web-research | Search plus main-content extraction with fail-closed robots checks, SSRF protection, redirect/size limits, and per-host pacing. |
ats-jobs | One typed model across public Greenhouse, Lever, Ashby, Workable, SmartRecruiters, and Recruitee boards. |
outcome-ledger | Durable deterministic idempotency keys and an honest |
coding-workflows | Unified-diff review, dependency-aware planning, bounded repo gates, traceback-to-file triage, and conventional commit synthesis. |
The complete machine-readable catalog is pinned at docs/manifest.json,
and the detailed protocol contract is in docs/API.md.
Grounded CV: demonstrable hybrid RAG
The corpus uses atomic evidence chunking: one independently verifiable fact per chunk, headings retained as metadata, no chunk crossing a section boundary. This makes every returned span independently citable and prevents a partially relevant paragraph from laundering an unsupported claim.
Two independent rankings run for every query:
Okapi BM25 over normalized terms and documented synonym expansion.
Cosine similarity over deterministic 384-dimensional dense feature-hash vectors built from words and character trigrams.
Reciprocal rank fusion combines both lists with RRF(k=60). The verifier then checks
claim-token coverage and numeric consistency against the retrieved spans. Supported
claims receive exact quotes; unsupported claims receive no citations.
Committed offline fixture, 10 queries, k=5:
Metric | Result |
recall@5 | 1.0000 |
MRR | 1.0000 |
nDCG@5 | 0.9920 |
Reproduce it with uv run grounded-cv-eval. The fixture is committed beside the profile,
and CI executes the same harness assertions without downloading a model or calling a
network service.
Trust and failure contract
Every tool has generated input and structured output JSON Schemas backed by Pydantic v2.
Every call has both an MCP-level deadline and an upstream HTTP timeout.
Public upstream failures become typed values (
timeout,blocked,not_found,rate_limited,upstream_error,parse_error, and others); low-level exceptions never cross a tool boundary.GitHub, web, and ATS text is explicitly marked
content_is_untrusted; prompts instruct clients not to execute instructions found inside it.Coding diffs, tracebacks, repository text, and gate output are untrusted. Quality gates use shell-free argv execution, credential-reduced environments, and hard deadlines; configured package-manager scripts should run only in repositories the caller trusts.
URL credentials, private networks, authenticated scraping, mutation APIs, and secret-shaped ledger fields are rejected. Optional
GITHUB_TOKENimproves GitHub rate limits but is never returned or logged.Tests run with outbound sockets disabled except loopback, require no keys, and use protocol clients plus committed provider fixtures.
The contract suite drives servers from
docs/manifest.json, and a discovery test derives the server list frompackages/on disk and fails when the two disagree. A server added without being declared would otherwise be skipped by every guard above while the suite still reported success.
Development
uv sync --all-packages --dev
uv run ruff check .
uv run ruff format --check .
uv run pytestCI runs those three gates on Python 3.11, 3.12, and 3.13. Licensed under MIT.
Available Tools
2 toolslookupC
Retrieve CV evidence with BM25+dense reciprocal-rank fusion.
Inputs: topic (required), limit. Returns: hits, query, status, failure, +4 more
verify_claimC
Verify claims and attach exact spans to each supported one.
Inputs: text (required). Returns: text, status, failure, sources, +3 more
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 2 tools
The two tools have clearly distinct purposes: lookup retrieves evidence, while verify_claim checks claims against that evidence. There is no ambiguity or overlap in their roles.
Both tool names are imperative verbs, but 'lookup' is a single word while 'verify_claim' follows a verb_noun pattern. This is a minor deviation, and the names are still predictable and readable.
With only two tools, the set feels slightly thin for a production-oriented server. However, the narrow scope of evidence retrieval and verification may justify the small number, making it borderline appropriate.
The two tools form a logical pipeline: retrieve evidence first, then verify claims against it. Minor gaps might exist, such as a way to browse or filter evidence sources, but the core workflow appears covered.
Maintenance
Related MCP Connectors
MCP server for your apps' tools and custom tools, plus hosted AI agents and approval-gated workflows
Hosted MCP server and Agent Skills for specs, engineering threads, and project memory.
A hosted MCP suite providing workflow primitives: web search, content fetching, compression, deduplication, cost/budget management, and agent memory pipelines.
Control plane for autonomous software labor. Agents claim objectives over MCP with audit trail.
Related MCP Servers
- FlicenseNot gradedqualityFmaintenanceMCP server that gives any LLM a managed Docker workspace with live browser, terminal, code execution, document skills, and autonomous sub-agents.126-
- FlicenseNot gradedqualityDmaintenanceA production-grade MCP server with 6 sandboxed tools and an agent orchestration engine for autonomous task completion, featuring an evaluation suite with CI/CD quality gates.-
- AlicenseCqualityCmaintenanceAn MCP execution layer that lets capable AI models safely control a developer workstation for file edits, terminal and process execution, Git operations, language-server analysis, browser and Windows UI automation, media processing, job/resource control, and checkpointing. It also coordinates durable single- or multi-agent tasks across providers with leases, dependency graphs, evidence logs, repair loops, and separate verification gates.1281ISC
- AlicenseNot gradedqualityAmaintenanceProvides a provider-agnostic MCP server that coordinates structured multi-agent workflows, owning task claims, action logs, handoffs, and server-side health gates for coding agents. It enables tracked specs, tasks, reviews, and verified completion across Claude Code, Codex CLI, Cursor, Grok Build, and OpenCode.29 npmMIT