Skip to main content
Glama

library-enrichment

A local-first library-evidence service for Rust and Python, exposed over MCP.

It answers the question a package index cannot: given a design objective and a concrete dependency environment, what relevant library capabilities exist, what evidence supports their use, what configuration do they require, and what remains unverified?

The service produces evidence. The calling agent produces judgments. It complements Context7 — Context7 explains concepts, this service establishes exact-release facts about the dependencies a project actually resolved.

Status

Phase 0 of 6. See STATUS.md for the current gate tally and docs/reports/acceptance.md for per-gate results. This repository is under active construction; the governing specification is complete and frozen, the implementation is not.

Related MCP server: mentu-navigator-mcp

What it does

Nine MCP tools over a shared evidence core:

Tool

Purpose

resolve_library

Establish exact release identity and environment before research

library_overview

Discover unfamiliar capabilities without knowing symbol names

search_evidence

Search a bounded set of API, docs, example, source, or release evidence

inspect_symbol

Characterize a candidate and its deployment requirements

compare_releases

Discover additions, removals, and non-API changes

verify_usage

Test a proposed invocation in an isolated capsule

read_artifact

Retrieve large result sections without flooding context

job_control

Observe or cancel a long-running operation

service_status

Inspect readiness and capabilities

Evidence carries provenance: every material fact traces to an exact artifact, a producer run, and a recorded environment. Declared availability, configured availability, type-check success, and runtime behavior are separate states — the service never collapses them.

Architecture

A Rust daemon owns identity, resolution, fetching, normalization, evidence storage, querying, job state, policy, and publication. Evidence lands in immutable Arrow/Parquet snapshots queried through DataFusion. A thin FastMCP 4 adapter exposes the tools; a separate worker runs Griffe extraction and isolated runtime probes.

coding agent ──┬── Context7 MCP ─────────► concepts, documentation examples
               │
               ├── library-research skill ► routing, evidence policy, synthesis
               │
               └── FastMCP 4 stdio adapter
                            │ typed, bounded local RPC
                            ▼
                   library-enrichmentd ──► fetchers, Rust/Python API producers,
                            │              LSP sessions, sandboxed verification
                            ▼
                   immutable evidence snapshots (Arrow/Parquet + DataFusion)

Service code, tool environments, extraction workspaces, caches, and outputs live outside every working repository. The repository under study is never a subprocess working directory, an extraction destination, or an install target.

Documentation

Path

Contents

docs/blueprint/

The governing specification — frozen, not edited

docs/architecture/

Compatibility matrix and design notes

docs/adr/

Decision records, including every deviation from the blueprint

docs/operations/

Install, configure, and run

contracts/

Frozen response-envelope schema and fixtures

tests/ACCEPTANCE_PLAN.md

The 46 acceptance gates

skills/library-research/

The companion agent skill

AGENTS.md

Instructions for agents working in this repository

Development

just --list        # every recipe
just doctor        # verify the toolchain and required producers are present
just ci            # gates for completed phases, plus the next one

Requires a pinned stable Rust toolchain (rust-toolchain.toml), uv, and just. Build and runtime verification profiles additionally require a working container runtime or bwrap.

License

Dual-licensed under either MIT or Apache-2.0, at your option.

Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in this work by you, as defined in the Apache-2.0 license, shall be dual licensed as above, without any additional terms or conditions.

Available Tools

9 tools
compare_releasesCompare ReleasesC
Idempotent

Discover additions/removals and non-API changes.

Inputs: name, cursor, scopes, ecosystem, +9 more. Returns: job, data, error, status, +10 more

inspect_symbolInspect SymbolA

Read retained symbol evidence; explicit execution options can run isolated semantic or runtime inspection.

Inputs: context_id (required), symbol_path (required), execution, max_bytes, +3 more. Returns: job, data, error, status, +10 more

job_controlJob ControlA

Observe or cancel an explicitly submitted long-running operation.

Inputs: job_id (required), action, max_bytes, wait_seconds, +1 more. Returns: job, data, error, status, +10 more

library_overviewLibrary OverviewB
Read-onlyIdempotent

Discover unfamiliar capabilities without knowing symbol names.

Inputs: context_id (required), area, discovery, max_bytes, +2 more. Returns: job, data, error, status, +10 more

read_artifactRead ArtifactA
Read-onlyIdempotent

Retrieve large result sections without flooding context.

Inputs: artifact_id (required), cursor, section, max_bytes. Returns: job, data, error, status, +10 more

resolve_libraryResolve LibraryC
Idempotent

Establish exact identity and environment before research.

Inputs: name (required), ecosystem (required), mode, extras, +12 more. Returns: job, data, error, status, +10 more

search_evidenceSearch EvidenceB
Read-onlyIdempotent

Search a bounded set of API, docs, examples, source, or release evidence.

Inputs: query (required), context_id (required), area, kinds, +4 more. Returns: job, data, error, status, +10 more

service_statusService StatusB
Read-onlyIdempotent

Inspect readiness and capabilities, without indexing.

Inputs: component. Returns: job, data, error, status, +10 more

verify_usageVerify UsageA

Test a proposed invocation or composition in isolation.

Inputs: snippet (required), context_id (required), mode, profile, +3 more. Returns: job, data, error, status, +10 more

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    B3.4/5.0

    Scored across 9 tools

    Disambiguation5/5

    Each tool maps to a distinct phase in the research workflow: status, identity, overview, evidence search, symbol inspection, release comparison, usage verification, artifact retrieval, and job control. Even the two preflight tools are clearly separated by readiness vs. identity.

    Naming Consistency3/5

    The majority use verb_noun (search_evidence, inspect_symbol, compare_releases, verify_usage, read_artifact), but service_status, library_overview, and job_control are noun_noun, creating an inconsistent pattern. Names remain readable, so not chaotic.

    Tool Count5/5

    Nine tools is a well-scoped count for a specialized library-research server; each tool covers a distinct step without redundancy or bloat.

    Completeness4/5

    The set covers the full discovery, inspection, comparison, verification, and retrieval lifecycle, with job control for long operations. The only minor gap is the absence of an explicit write-back/enrichment-submission tool, though the server appears read-oriented by design.

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      D
      maintenance
      Provides up-to-date, version-specific documentation and code examples for software libraries directly to LLMs, enabling resolution of library identifiers and retrieval of relevant documentation with code snippets.
      MIT
    • A
      license
      Not graded
      quality
      C
      maintenance
      Provides read-only, provenance-first repository navigation for agents and humans, with ranked lexical retrieval, exact query, document handles, symbol context, and change impact analysis.
      64 npm
      Apache 2.0
    • F
      license
      Not graded
      quality
      B
      maintenance
      Enables AI coding agents to interrogate a local repository deterministically for symbols, dependencies, call paths, routes, tests, and Git impact, returning verified structured evidence with exact line citations instead of requiring the whole codebase to be read.
      261 PyPI
      4
      -