@draftboard/mcp

⚠️ Superseded — use the hosted server instead

Draftboard now runs the MCP server for you at https://mcp.draftboard.com. No install, no Node, no API key in a config file — you approve it in the browser like any other connected app, and it stays up to date on its own. Setup for every client: ****

This package still works. Nothing here has been switched off and nothing is about to break, so there is no rush — but it will not get new tools, and it is no longer what we point people at.

Moving

# Claude Code
claude mcp add --transport http draftboard https://mcp.draftboard.com
claude mcp login draftboard

# Codex CLI
codex mcp add draftboard --url https://mcp.draftboard.com
codex mcp login draftboard

Claude Desktop: Settings → Connectors → Add → Add custom connector → https://mcp.draftboard.com

Then remove the old draftboard entry from your MCP config, and delete the API key you created for it (Settings → API keys) — the hosted connection issues and revokes its own.

One difference worth knowing before you move

The hosted server is approved in a browser instead of reading DRAFTBOARD_API_KEY from the environment. That means you have to be signed in to Draftboard in a browser on the machine doing the approving. If you run this headlessly — CI, a server, a container with no browser — stay here for now; that case is not yet covered.

The tools are the same. Both surfaces expose the identical set, checked by a test that compares them, so moving is not a trade-off in what your assistant can do.

A Model Context Protocol server for the Draftboard Integration API. It lets any MCP-capable agent (Claude, Codex, etc.) work with your Draftboard warm-introduction data — targets, connection paths, ranks, and tags — to answer questions like “who are my best intro opportunities right now?” or “am I already connected to these people?”.

Draftboard maps how your team is really connected to your prospects; this server exposes that data over MCP. API reference: .

Just want to get intros (not build on the API)? Also install the companion draftboard-intros skill — it teaches your assistant which tool to reach for and how to read the results. This server is only the engine.

Requirements

• Node.js 20+
• A Draftboard Integration API key (Pro/Team plan with API access). Find it in the Draftboard app under Settings → API keys.

Quick start

The server speaks MCP over stdio. Point your client at it and pass the key via env.

Claude Code / Claude Desktop (.mcp.json or claude_desktop_config.json)

{
  "mcpServers": {
    "draftboard": {
      "command": "npx",
      "args": ["-y", "github:draftboardco/mcp"],
      "env": {
        "DRAFTBOARD_API_KEY": "db-api_xxxxxxxx"
      }
    }
  }
}

Codex CLI

codex mcp add draftboard --env DRAFTBOARD_API_KEY=db-api_xxxx -- npx -y github:draftboardco/mcp

or in ~/.codex/config.toml:

[mcp_servers.draftboard]
command = "npx"
args = ["-y", "github:draftboardco/mcp"]
startup_timeout_sec = 60   # first `npx github:` launch clones + builds the server

[mcp_servers.draftboard.env]
DRAFTBOARD_API_KEY = "db-api_xxxxxxxx"

Run from source

npm install
npm run build
DRAFTBOARD_API_KEY=db-api_xxxx node dist/index.js

Verify your key

DRAFTBOARD_API_KEY=db-api_xxxx npm run smoke         # checks get_me
DRAFTBOARD_API_KEY=db-api_xxxx npm run smoke -- --full  # also runs a status overview

Configuration

The key is used only to call Draftboard’s API from your machine — it is never sent to the model provider (OpenAI / Anthropic) or any third party, and the server never logs the Authorization header.

Tools

Thin tools (1:1 with the API, raw JSON):

Extended tools (rest of the API; ⚠ = changes data, host-approved at runtime):

Stars = the rating, and nothing else. The star glyphs describe the rating only. tier is the raw wire spelling of the same setting (1–5, lower is better) and never carries them, and the legacy preferred flag below is a separate boolean, not a rating at all.

Four capabilities, kept separate:

Legacy toggles (⚠ still wired, still work — the product moved both onto the rating, so prefer set_connector_tier for new work):

• set_connector_preferred — mark/unmark a connector as a preferred supporter. preferred is its own boolean column and not the rating: set_connector_tier never writes it, so rating: 5 does not mark someone preferred, and marking someone preferred does not give them a rating.
• set_connector_excluded — exclude/un-exclude a connector from warm-path results. rating: 1 hides a connector and sets this flag for you, but the sync runs one way: excluded: false does not clear a rating: 1. To un-hide someone, give them a rating of 2–5.

Prospecting tools (company-first discovery — ⚠ BETA · Team/Enterprise · async):

Outcome tools (composed, mapped to real jobs):

Outcome tools that walk connections are bounded and return a telemetry block (targetsMatched, targetsScanned, connectionsFetched, truncated, nextSuggestedFilter) so you always know the coverage of an answer. Scope them with filters before running on large lists.

Finding one person: use resolve_target (or check_if_connected for a batch). Both look each URL up directly, so they answer in about a second on a book of any size. Never page list_targets hunting for someone — that is one request per 100 targets, and list_targets does not return targets whose paths have not been computed yet, so they would look missing when they are not.

Setup (for your assistant)

If you’re an AI assistant that a user asked to set up Draftboard from this repo, do this, then tell the user it’s ready:

0. Check npx --version (needs Node.js 20+). If it’s missing, point the user to to install Node and stop until they have.
1. Configure a stdio MCP server named draftboard in the user’s client: command npx, args ["-y", "github:draftboardco/mcp"], env DRAFTBOARD_API_KEY = the key the user gave you. On Claude, use your MCP-add command or edit .mcp.json / the Claude Desktop config; on Codex CLI, run codex mcp add draftboard --env DRAFTBOARD_API_KEY= -- npx -y github:draftboardco/mcp. Never print the key back.
2. Also install the companion skill so the assistant uses these tools well: copy the draftboard-intros/ folder from draftboardco/draftboard-skill into the user’s skills directory — ~/.claude/skills/ for Claude Code, ~/.codex/skills/ for Codex CLI.
3. Verify with the get_me tool — a returned name means it works. Missing tools → restart the client; 401 → the key is wrong or expired.

Feedback & support

Early beta — please report issues at github.com/draftboardco/mcp/issues. The server never logs your API key, so pasting the full error output is safe and speeds up a fix.

License

MIT