pipedrive-mcp

Pipedrive CRM as MCP tools. Find deals, people and companies, write to them, and move deals through a pipeline.

A single Go binary that speaks Model Context Protocol over stdio — protocol revisions 2026-07-28 (the current one) back to 2024-11-05, negotiated per client. It runs as a subprocess of your client, on your own machine, against one Pipedrive workspace: search by name and get back the ids everything else needs, read deals, people, organizations, activities and notes with their custom fields under the names your workspace gives them and dropdown values under their labels, create and edit any of them, and move a deal through its pipeline — won, lost, reopened, or into another stage.

Why pipedrive-mcp

Pipedrive's own API is two APIs: a v2 that is current and a v1 whose sunset date, 2026-07-31, has passed. This server commits to v2 and carves out only what exists nowhere else — notes, and the whoami lookup. Those four tools run on an API that is out of support; they still answer, and there is no v2 to move them to, because v2 exposes no /notes and no /users. Both carve-outs are documented at their call sites and in CHANGELOG.md, because a v1 dependency nobody wrote down is one nobody migrates — and a 410 from either now reaches you as [gone], naming the sunset, rather than as a validation error blaming your request.

Its tool surface follows the shipped Google Workspace MCP servers rather than a house style of its own: reads stay discrete, every mutation goes through one manage_ tool with an action, and writes explain their refusals. Those are the servers a model has most likely already seen, so matching them is what makes this one legible without being explained.

Writing is guarded because Pipedrive has no undo. Every write reads its target first and refuses to replace a field that already holds a value unless you say so, naming each field it is protecting. A refusal you cannot act on is a bug.

Every tool below is driven against a real Pipedrive workspace as well as against the fakes the tests use — reads, resources, the guard refusals, the rehearsal paths, and a reversible write on each resource that has one. That rundown is a Go suite rather than a habit now: it lives in internal/integration/ behind -tags=integration, and make integration runs it. Two things still bound it. The write probes need a second opt-in on top of the tag, PIPEDRIVE_INTEGRATION_WRITES=1, so a run without that proves the read half only — writing to a live CRM is a decision, not a default. And it has been exercised against one workspace, so a custom-field or permission setup unlike that one is untested ground.

Install

go install github.com/mmedum/pipedrive-mcp/cmd/pipedrive-mcp@latest

That puts the binary in Go's bin directory, which is often not on your PATH. If the next command says command not found, either use the full path or add the directory once:

"$(go env GOPATH)/bin/pipedrive-mcp" --version    # check it landed
export PATH="$(go env GOPATH)/bin:$PATH"          # or add it to your shell profile

Or take a signed archive from the latest release — Linux, macOS and Windows, on amd64 and arm64 — and put the binary on your PATH. Nothing about a release has to be taken on trust:

# --ignore-missing, because SHA256SUMS covers every archive and every
# SBOM, and you will have downloaded one of them.
sha256sum -c SHA256SUMS --ignore-missing

# One signature over the checksum file, keyless, tied to the release
# workflow's own identity.
cosign verify-blob SHA256SUMS \
  --bundle SHA256SUMS.bundle \
  --certificate-identity-regexp 'https://github.com/mmedum/pipedrive-mcp/.*' \
  --certificate-oidc-issuer 'https://token.actions.githubusercontent.com'

# And the archive itself carries build provenance.
gh attestation verify pipedrive-mcp-X.Y.Z-linux-amd64.tar.gz --owner mmedum

Set up Pipedrive

The server stores your API token in the OS keyring (libsecret on Linux, Keychain on macOS, Credential Manager on Windows) rather than asking you to paste it into a JSON config. Full reference and validation rules: docs/configuration.md.

pipedrive-mcp login
# Pipedrive workspace subdomain (e.g. acme for acme.pipedrive.com): your-subdomain
# Pipedrive API token for "your-subdomain": ········
# login: token stored in OS keyring (service=pipedrive-mcp, account=your-subdomain)
# login: default domain recorded in ~/.config/pipedrive-mcp/config.json

Pass --domain your-subdomain (or set PIPEDRIVE_COMPANY_DOMAIN) to skip the domain prompt — useful in scripts.

After that, your MCP client config needs nothing about the workspace domain or the token — both are resolved from the keyring + a small non-secret pointer file (os.UserConfigDir()/pipedrive-mcp/config.json). Run pipedrive-mcp status to verify, or pipedrive-mcp logout to clear them.

To override the recorded domain in one shell (e.g. point at a different workspace temporarily), set PIPEDRIVE_COMPANY_DOMAIN — env wins over the userconfig pointer. For CI, automation, or one-off use of a different token, set PIPEDRIVE_API_TOKEN. Env takes precedence over the keyring, matching the gh and aws CLIs.

Connect a client

Claude Desktop

After pipedrive-mcp login has stored the token in your keyring AND recorded the default domain in user-config, the Claude Desktop config needs nothing — no secret, no domain:

{
  "mcpServers": {
    "pipedrive": {
      "command": "/usr/local/bin/pipedrive-mcp"
    }
  }
}

To pin a specific workspace (e.g., when you have several stored), add PIPEDRIVE_COMPANY_DOMAIN under env to override the recorded default for this MCP server only.

Configuration

Every setting is a PIPEDRIVE_* environment variable. The full list, with defaults and validation rules, is in docs/configuration.md. The ones that change what the server will do at all:

Tools

Twenty tools. Reads stay discrete and every mutation goes through one manage_ tool with an action, which is how the Google Workspace MCP servers shape theirs.

Start with search. It is the natural-language gateway: it turns a name into the numeric id every other tool needs. Reaching for list_deals to find "the Acme deal" is the common mistake — it filters, it does not match names.