|

MCP Server

Let AI agents operate Cencori via the Model Context Protocol — reads, inference, and safe manual-action guidance.

MCP Server

@cencori/mcp is Cencori's official Model Context Protocol server. It lets AI agents in Cursor, Claude Desktop, and any MCP client work with Cencori — search docs, view metrics, read memory/agents/sessions/governance, and run multimodal inference.

Quick start

 
npx @cencori/mcp

Docs and manual-action guidance tools work with no API key. Add CENCORI_API_KEY to enable platform reads, and CENCORI_MCP_WRITE=1 to enable inference.

Client config

{
  "mcpServers": {
    "cencori": {
      "command": "npx",
      "args": ["-y", "@cencori/mcp"],
      "env": {
        "CENCORI_API_KEY": "csk_your_project_key",
        "CENCORI_MCP_WRITE": "1"
      }
    }
  }
}

Add this to ~/.cursor/mcp.json, your Claude Desktop config, or any MCP client. Restart the client after changing env vars.

Action tiers

The server is safe by default — anything that costs money or changes state is opt-in.

TierGateTools
ReadCENCORI_API_KEYlist_models, get_metrics, get_health, check_quota; agents, memory, sessions, and governance list/get tools
WriteCENCORI_MCP_WRITE=1Inference (generate_text, generate_rag, create_embeddings, moderate_content, generate_image, vision, documents); memory (remember_memory, write_memory, create_namespace); agents (create_agent, update_agent); sessions (create_session, add_session_turn); governance drafts (create_policy, install_template)
DestructiveCENCORI_MCP_DESTRUCTIVE=1delete_memory, delete_agent, delete_session, approve_session, reject_session (implies write)
Publicnonesearch_docs, get_doc, list_docs, get_integration_guide (the llm.txt setup contract), and how_to_* guidance

Manual-only actions

Some actions are never performed by the MCP — the agent tells you how to do them yourself. Each how_to_* tool returns step-by-step instructions plus the exact dashboard link:

  • API keys — how_to_create_api_key, how_to_edit_api_key, how_to_revoke_api_key
  • Governance (checker steps) — how_to_activate_policy, how_to_respond_to_change_request
  • Billing — how_to_change_plan, how_to_manage_billing
  • Access — how_to_manage_members

This keeps credential, billing, and access decisions with a human, even when an agent is driving.

Environment variables
VariableDefaultDescription
CENCORI_API_KEY—Project key (csk_...) for platform tools
CENCORI_MCP_WRITEfalseEnable inference tools
CENCORI_MCP_DESTRUCTIVEfalseEnable destructive actions (implies write)
CENCORI_MCP_FEATURESallComma list: docs, guidance, gateway, agents, memory, sessions, governance, multimodal
CENCORI_BASE_URLhttps://cencori.comPlatform API host

Security

The server is a thin adapter over the Cencori API — every call carries your key and the platform enforces project scope, tier limits, and governance rules. The MCP adds a second, client-side guardrail: reads are always on, writes and destructive actions require explicit env flags, and credential/billing/access actions are guidance-only.

Did you like the content?