MCP Server

Context Chain exposes its knowledge graph through an MCP server with 9 tools. Any MCP-compatible coding AI can query design decisions while writing code.

Tools

get_context_for_code

The main entry point. Five-slot retrieval for a piece of code.

Parameter	Type	Description
file_path	string	Current file being edited
function_name	string	(Optional) Specific function
code_snippet	string	(Optional) Code fragment for semantic matching

Queries all five slots in priority order: code anchor → keywords → decision links → semantic fallback, with metadata filtering throughout. Returns ranked decision summaries; request full content with a follow-up call.

search_decisions_by_keyword

Direct keyword search against the inverted index. Catches business terms ("refund", "auth", "rate limiting") that don't appear in code.

Parameter	Type	Description
keywords	string[]	Terms to search
repo	string	(Optional) Scope to a repo

search_decisions_semantic

Vector similarity search. Useful when the decision's wording differs from the search — e.g., searching "prevent duplicate charges" finds decisions about "idempotency design".

Parameter	Type	Description
query	string	Natural language query
limit	number	(Optional) Max results

get_decision_relationships

Explore causal, dependency, and conflict chains from a decision.

Parameter	Type	Description
decision_id	string	Starting decision
depth	number	(Optional) How many hops to follow

get_code_structure

List functions in a file or service.

Parameter	Type	Description
path	string	File path or service name
type	string	Filter: function, file, service

get_callers

Upstream dependencies — who calls a given function.

Parameter	Type	Description
function_name	string	Function to look up
repo	string	Repository scope

get_callees

Downstream dependencies — what a function calls.

Parameter	Type	Description
function_name	string	Function to look up
repo	string	Repository scope

get_cross_repo_dependencies

Cross-service API dependencies. Shows how services connect through API calls, shared packages, and message queues.

Parameter	Type	Description
repo	string	Repository to check

report_context_usage

Feedback loop — report which decisions were actually useful during a coding session. Improves retrieval ranking over time.

Parameter	Type	Description
decision_ids	string[]	Decisions that were used
session_context	string	(Optional) What you were working on

Starting the Server

npm run mcp

Communicates over stdio, as expected by MCP clients. Usually started automatically by Claude Code via .mcp.json.

Connection

See Claude Code Setup for configuration.