MCP servers that ship as code — a repository or a package you install and run locally (usually over stdio). They have no hosted endpoint, so we cannot probe or grade them: instead each one shows its repository, install command, license, stars and when it was last updated.
Collected from GitHub repositories and npm, PyPI, Cargo, Go, Maven, NuGet and RubyGems packages tagged as MCP servers, the official MCP Registry, Docker's MCP catalog, ModelScope, Baidu MCP World, Cline and other public directories; one entry per repository. Each candidate is checked by a language model from its name, description and topics, and only actual MCP servers are listed (not clients, SDKs, lists or skills). Package metadata comes from ecosyste.ms (CC BY-SA 4.0). A server that also offers a hosted endpoint is listed under Remote instead. These entries are not counted in our reachability, verification or readiness statistics.
-
mcp-confidencegithub.com/shaxzodbek-uzb/mcp-confidenceupdated this year
A drop-in confidence gate for LLM agents: turn token logprobs into an accept / verify / ask-a-human routing decision, with honest per-model calibration. Local & open-model first, plus an MCP server.
MIT -
verifydocgithub.com/bhaskargurram-ai/verifydocupdated this year
Trust layer for document→JSON extraction & AI agents: calibrated per-field confidence + source grounding + accept/review abstention on any OCR/VLM. Ships VerifyDocBench, a novel grounding-conditioned conformal method, and an MCP server.
Apache-2.0 ★ 1 -
Compare classifier confidence with accuracy using your own predictions. Browser-based calibration analysis and reports.
npm MIT -
Per-repo memory, outcome telemetry, and a calibrated-confidence gate for Claude Code, with MCP and AGENTS.md projections so other AI coding tools can read its context. Notes survive sessions; success claims need test evidence; your reverts are remembered. Local-only, stdlib runtime.
Apache-2.0 ★ 9 -
Cheap, confidence-scored question answering over knowledge graphs: each hop is a calibrated classification decision. Library, CLI, and MCP server.
pypi Apache-2.0 -
MCP server for Laya typed decisions (noul / choice / score): warm model sidecar, token-budget preflight, persisted calibration, honest confidence.
Apache-2.0 ★ 5 -
MCP server that triages a Jira board with typed, calibrated decisions from a small local model — confidence, human-review flags, held-out evaluation. Synthetic bank-themed data.
-
? Tiny Decision - A Local Harness & Deployment Platform for System One Models. In-process GGUF/ONNX runtime, calibrated confidence, visual DAG pipelines, and MCP export.
MIT ★ 1 -
Typed answers with calibrated confidence for agents. Local-first, no API key.
pypi -
Epistemic Status Tracking MCP server — tracks what an agent KNOWS vs INFERS vs ASSUMES with calibrated confidence.
Apache-2.0 -
A Rust System-1 decision layer: typed decisions with calibrated confidence over MCP stdio, feeding deterministic DAG route nodes
MIT -
olaverse-mcpactive
Olaverse Lab's API as MCP tools: document extraction with calibrated confidence, decisions, language ID (608 languages), diacritics and question generation
pypi Apache-2.0 -
LangGraph after-sales agent that cites evidence, refuses when evidence is weak, and learns from human review. Hybrid RAG + reranker, calibrated confidence gate, MCP tools, live replay.
MIT -
A Jev-compatible decision layer as an MCP server: typed noul/choice/score questions with probability distributions and confidence. Pluggable backends (local needle/laya models or any OpenAI-compatible endpoint), calibration tracking, conditions v1, eval harness.
MIT -
The context quality layer for AI agents — memory that checks itself: lifecycle governance, calibrated confidence, and staged claim gates. Local-first, MCP 13 tools, DeepSeek Harness plugin.
npm Apache-2.0 ★ 12 -
Memory that abstains instead of guessing: agent memory in a local SQLite file or your own Postgres, with a verdict, confidence and provenance on every hit, and a calibrated refusal when nothing clears the threshold.
pypi AGPL-3.0 ★ 7 -
MCP server that lets Claude Code, Codex & AI agents drive a real Chrome browser with JEV (TypeSafe System One): ~100 ms typed decisions, calibrated confidence, human-in-the-loop questions, safety gates and a debug trace.
MIT ★ 7 -
Jev MCP server: a decision layer for coding agents, built on TypeSafe Jev (System One model). Ship gates, risk checks, file triage that keeps files out of context, and a safe headless browser, with calibrated confidence. For Claude Code, Codex, Cursor.
npm MIT ★ 6 -
Squish gives AI agents calibrated recall confidence (ECE 0.055) and honest abstention — the only memory runtime that tells your agent how much to trust each recollection, and says I don't know when nothing clears the bar. Local-first, MIT, npm install in 30s.
MIT ★ 12 -
Occupation classification and job-title normalization for the O*NET-SOC taxonomy. A fine-tuned retrieve-and-rerank model maps job titles, resumes, postings and skill lists to ranked SOC codes with calibrated confidence. 51% top-1 vs 21% TF-IDF. Python, CLI, REST API, MCP server. Runs locally.
pypi MIT ★ 2 -
MCP server and agent skill for the TypeSafe AI System One API (Jev): decompose a judgment into Choice / Score / Noul questions, lint them, measure on labelled data, and put calibrated thresholds in code
MIT ★ 4 -
MCP server that compiles checkable task contracts, retrieves quote-grounded web evidence, and verifies drafts for coding agents — especially cheap/small host models.
pypi MIT -
ClearRun MCP server. Scores AI answers for honesty (confidence-justification alignment) through the hosted API. Not a fact-checker.
npm MIT