Skip to content

MCP Server

The MCP server exposes code-context-agent's core differentiators via the Model Context Protocol, enabling coding agents (Claude Code, Cursor, etc.) to use analysis capabilities directly.

Commodity tools (ripgrep search, LSP symbols, git history, ast-grep) are intentionally not exposed -- they are already available in the MCP marketplace. The server focuses on capabilities that are unique to code-context-agent.

Starting the Server

code-context-agent serve

Used for local MCP clients like Claude Code and Claude Desktop.

code-context-agent serve --transport http --port 8000

Used for networked access by remote MCP clients.

Tools

start_analysis

Kicks off a full codebase analysis. Returns immediately with a job_id for polling.

The analysis runs in the background (5--20 minutes) and produces the standard .code-context/ artifact set.

Parameter Type Default Description
repo_path str required Absolute path to the repository root
focus str "" Optional focus area (e.g., "authentication", "API layer")
issue str "" Optional GitHub issue reference (e.g., gh:1694)

check_analysis

Polls the status of a running analysis job. Call every 30 seconds until status is "completed" or "error".

Parameter Type Description
job_id str The job ID returned by start_analysis

Status values: starting, running, completed, stopped, error.

query_code_graph

Runs graph algorithms on a pre-built code graph. Requires .code-context/code_graph.json from a prior analysis.

Algorithm Description Requires
hotspots Betweenness centrality -- bottleneck/integration code --
foundations PageRank -- core infrastructure --
trust TrustRank -- noise-resistant importance scoring --
modules Louvain community detection -- logical clusters --
entry_points Nodes with no incoming edges --
coupling Connection strength between two nodes node_a, node_b
similar Personalized PageRank from a node node_a
dependencies BFS traversal of transitive dependencies node_a
triangles Tightly-coupled triads --
category Nodes in a business logic category category

explore_code_graph

Progressive graph exploration, starting broad and drilling down.

Action Description Requires
overview Entry points, hotspots, modules, foundations --
expand_node Neighbors and relationships of a node node_id
expand_module Internals of a detected module module_id
path Shortest path between two nodes node_id, target_node
category All nodes in a business logic category category
status Current exploration state --

get_graph_stats

Returns summary statistics (node/edge counts by type, density) for a repository's code graph.

Resources

The server provides read-only access to analysis artifacts via URI templates:

Resource URI Artifact
analysis://{repo_path}/context CONTEXT.md -- narrated architecture overview
analysis://{repo_path}/graph code_graph.json -- structural graph data
analysis://{repo_path}/manifest files.all.txt -- complete file listing
analysis://{repo_path}/signatures CONTEXT.signatures.md -- compressed signatures
analysis://{repo_path}/bundle CONTEXT.bundle.md -- curated source bundle
analysis://{repo_path}/result analysis_result.json -- structured analysis metadata

Configuration for Clients

Add to your .mcp.json:

{
  "mcpServers": {
    "code-context-agent": {
      "command": "code-context-agent",
      "args": ["serve"]
    }
  }
}

Add to your MCP configuration:

{
  "mcpServers": {
    "code-context-agent": {
      "command": "code-context-agent",
      "args": ["serve"],
      "transport": "stdio"
    }
  }
}

Typical Workflow

  1. Check if .code-context/code_graph.json exists in the target repository
  2. If not, run start_analysis(repo_path) and poll check_analysis(job_id) until done
  3. Use explore_code_graph(action="overview") for a high-level view
  4. Use query_code_graph(algorithm="hotspots") to find critical code
  5. Drill down with explore_code_graph(action="expand_node", node_id="...") on interesting results
  6. Read artifacts via resources for full content (e.g., analysis://repo/path/context)