All projects

Agent Infrastructure / Developer Tooling

current

Repo Context Forge

Local-first MCP repository intelligence and agent platform for deterministic, source-grounded code analysis within strict read-only boundaries.

View repository

01 / Overview

Why it exists

Repo Context Forge is a local-first MCP repository intelligence and agent platform with 40 configured tools across six local servers, plus a bounded source-grounded agent that uses a restricted tool subset by default.

Provides secure repository access, deterministic search and analysis, and reproducible context packs without executing analyzed code or relaxing filesystem, path, process and tool-argument boundaries.

02 / Architecture

System architecture

Typer and FastMCP adapters call explicit application factories that construct deterministic repository intelligence services. Those services read only bounded local repository mounts and write generated context packs to a separate output boundary. Analyzed code is not executed and repository content is not sent to a hosted language model.

Typer and FastMCP adapters call explicit application factories that construct deterministic repository intelligence services. Those services read only bounded local repository mounts and write generated context packs to a separate output boundary. Analyzed code is not executed and repository content is not sent to a hosted language model.

Bounded tool entry

Transport adapters remain thin; application factories construct the domain services and their security dependencies.

Typer CLI and MCP Clients adapters Application Factories. Application Factories construct Repository Intelligence Services.

  1. ClientTyper CLIlocal commands
    ClientMCP ClientsFastMCP tool adapters
  2. ControlApplication Factoriesexplicit dependencies
  3. AnalyzerRepository Intelligence Servicesdeterministic domain layer
    • Repository access
    • Code search
    • Python symbols
    • Dependency analysis
    • Read-only Git
    • Context generation

Read boundary to write boundary

Analyzed workspaces and generated artifacts are separated by containment checks and bounded analyzers.

Local Repositories bounded reads Restricted Analyzers. Restricted Analyzers atomic artifacts Context Output.

  1. BoundaryLocal Repositoriesindividually mounted · read only
  2. AnalyzerRestricted Analyzersno import · no execution
  3. OutputContext Outputseparate writable root

Architecture notes

  • Analyzed repositories are local, individually mounted read only and never executed or imported.
  • Git is the only subprocess boundary and is restricted to allowlisted, read-only commands.
  • Context output uses a separate writable root with containment checks and atomic replacement.
  • The platform configures 40 tools across six local MCP servers; its local Ollama agent uses a bounded read-only subset and does not send repository content to a hosted LLM.

03 / Engineering concepts

System concerns made explicit

  • Repository Intelligence
  • Source-Grounded Context
  • Deterministic Code Search
  • Symbol Analysis
  • References
  • Callers / Callees
  • Dependency Analysis
  • Dependency Graphs
  • Read-Only Git Intelligence
  • Context Packs
  • Task Bundles
  • Validated Tool Arguments
  • Filesystem / Path Containment
  • Read-Only Process Boundaries
  • Bounded Local Agent

04 / Proof & evidence

Measured or reproducible signals

Configured MCP surface

40 tools

Six local MCP servers

The platform configures 40 tools across six servers; the read-only local agent uses a bounded subset by default.

Prerelease validation

206 tests · 87% coverage

v0.2 alpha validation

Validation covers deterministic analyzers, tool arguments, context generation and filesystem, path, Git and process security boundaries.

05 / Technology stack

Current implementation

  • Python
  • MCP
  • FastMCP
  • Python AST
  • Typer
  • Docker
  • Git
  • Ollama