Skip to main content
Glama
graphite-mcp

graphite-mcp

Official
by GraphiteAI

graphite-mcp

MCP server for the Graphite Financial Knowledge Graph — wires the graph into Claude Code, Claude Desktop, Cursor, Codex CLI, and any other MCP-compatible client.

Your Claude / agent does the reasoning. Graphite answers the graph questions it asks for. No LLM tokens are billed by Graphite — you bring your own LLM (subscription or API key); we only serve graph data.

Install

pip install graphite-mcp

Or from source:

git clone https://github.com/GraphiteAI/graphite-mcp
cd graphite-mcp
pip install -e .

Related MCP server: Neo4j YASS MCP

Get a Graphite key

Free tier — 100 graph queries / month — at: https://graph.graphite-ai.net/#/portal

Sign up with your email; you get back a key like sk-….

Chat-with-the-graph in the web portal is BYOK — you bring your own Anthropic or OpenAI key (it stays in your browser). Or skip the portal and use this MCP server with your Claude / ChatGPT subscription, no API key needed on the LLM side at all.

Configure your client

Same JSON shape works for every MCP client; only the file path differs.

{
  "mcpServers": {
    "graphite": {
      "command": "graphite-mcp",
      "env": {
        "CENTRAL_SERVER_URL": "https://api.graphite-ai.net",
        "CUSTOMER_API_KEY": "sk-your-graphite-key"
      }
    }
  }
}

Client

Config path

Claude Code

~/.claude/mcp.json

Claude Desktop

~/Library/Application Support/Claude/claude_desktop_config.json (macOS)

Cursor

~/.cursor/mcp.json

Codex CLI

~/.codex/config.json

Restart your client after editing.

If you installed from source and don't want to put the package on PATH, replace the command line with:

"command": "python",
"args": ["-m", "graphite_mcp"],

Available tools

Once configured, Claude can call these tools to answer questions about companies, supply chains, executives, regulations, and patents.

Tool

Purpose

search_entities

Free-text search by name, ticker, sector, or description

get_entity

Full record for one entity by ID (e.g. company:NVDA)

get_relationships

Every edge attached to an entity (employs, supplies, depends_on, …)

get_facts

Typed fact log for an entity (revenue, headcount, etc.)

find_path

Shortest path between two entities through the graph

exposure_analysis

1st + 2nd-degree neighborhood, sector breakdown — supply-chain risk view

compare_entities

Shared connections + path distance + direct relationships between two

Example prompts

After setup, just ask Claude:

  • "What's NVIDIA's supply-chain exposure to TSMC?"

  • "Find the shortest path from company:AAPL to company:ASML."

  • "Compare Microsoft and Google — who do they share board members with?"

  • "List every revenue-from edge for NVDA."

Claude will pick the right tool, call this MCP server, and ground its answer in real graph data.

How it works

your Claude          graphite-mcp           api.graphite-ai.net
─────────────       ────────────────       ─────────────────────
"NVDA exposure?" ─→ exposure_analysis ─→  GET /graph/exposure
                    (this package)         (returns JSON)
                                       ←─
              ←─    formatted answer

The MCP server is a thin stdio adapter — it doesn't store anything, doesn't see your prompts, doesn't bill you. All the value lives in the graph at api.graphite-ai.net.

Env vars

Var

Default

Notes

CENTRAL_SERVER_URL

http://localhost:8000

Override to use a different Graphite deployment

CUSTOMER_API_KEY

(empty)

Required — issued at https://graph.graphite-ai.net/#/portal

License

MIT. Use freely, modify freely, no warranty.

Available Tools

7 tools
compare_entitiesA

Compare two entities: shared connections, direct relationships, path distance. Example: compare_entities(entity_a='company:NVDA', entity_b='company:AMD')

Inputs: entity_a (required), entity_b (required).

exposure_analysisA

Analyze a company's exposure: 1st and 2nd degree connections, sector concentration, dependency risks. Example: exposure_analysis(entity_id='company:TSM')

Inputs: entity_id (required).

find_pathA

Find how two companies/entities are connected through the knowledge graph. Shows the chain of relationships. Example: find_path(source='company:AAPL', target='company:TSM')

Inputs: source (required), target (required), max_depth.

get_entityA

Get detailed information about a specific entity by ID. Example: get_entity(entity_id='company:NVDA')

Inputs: entity_id (required).

get_factsA

Get known facts about an entity — revenue, employee count, etc. Example: get_facts(entity_id='company:AAPL')

Inputs: entity_id (required).

get_relationshipsB

Get all relationships for an entity — suppliers, competitors, partners, dependencies, etc. Example: get_relationships(entity_id='company:NVDA')

Inputs: entity_id (required).

search_entitiesB

Search the financial knowledge graph for companies, people, patents by name, ticker, or description. Example: search_entities(query='NVIDIA') or search_entities(query='semiconductor', sector='semiconductors')

Inputs: query (required), entity_type, sector, limit.

Tool Schema Changelog

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

Loading tool schema changelog…

    TDQS

    A4/5.0

    Scored across 7 tools

    Disambiguation5/5

    Each tool has a clearly distinct purpose: search, detail retrieval, relationship listing, fact retrieval, path finding, exposure analysis, and pairwise comparison. No two tools appear to do the same thing, and descriptions specify unique parameters and outputs. Even the two-entity tools (find_path and compare_entities) are differentiated by their focus on connection chains versus shared relationships.

    Naming Consistency4/5

    Most tools follow a consistent snake_case verb_noun or verb_noun pattern (search_entities, get_entity, get_relationships, get_facts, find_path, compare_entities). The only slight deviation is exposure_analysis, which uses a noun phrase rather than a verb-prefixed form, but this is minor and does not create confusion.

    Tool Count5/5

    Seven tools is an appropriate size for a specialized financial knowledge graph server. Each tool addresses a distinct query need without redundancy or bloat. This falls well within the ideal 3-15 range for a focused domain.

    Completeness5/5

    The tool surface covers the core lifecycle of knowledge graph exploration: search, retrieve entity details, list relationships, retrieve facts, trace paths, analyze exposure, and compare entities. There are no obvious missing operations for a read-only analytical server. All tools can be chained since they accept entity IDs returned by search or get operations.

    Maintenance

    ActivityInactive
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers