callrail-mcp

A Model Context Protocol server for the CallRail REST API v3. Point Claude Code, Claude Desktop, Cursor, or any other MCP client at it and ask about your call tracking in plain English.

I'm Steve Japalucci. I run Pittsburgh Digital Marketing Agency (PGHDMA), and I built this for my own client work. CallRail bills per tracking number and per minute across every client I manage, and I was tired of clicking through its reports to answer questions like "which client is burning my minutes this month". Now I ask Claude.

Works with both major versions of the MCP Python SDK (1.x and 2.x).

What I use it for

These are the kinds of questions I ask it every week:

Reporting

• "Pull last week's calls for Alan Construction, grouped by source"
• "Show me every missed call this month"
• "Find any calls from 412-555-1234 across all clients in the last 90 days"
• "Get the transcript for call CAL019abc..."

Agency cost attribution

• "Why is my CallRail bill $174? Break it down by client"
• "Which client is the biggest minute user this cycle?"

Conversion debugging

• "Why didn't this call convert in Google Ads? CAL019..."
• "Is this 58-second call eligible to count as a Google Ads conversion?"

Lead management

• "Show me everything this person has ever done: calls, forms, texts"
• "Tag this SMS thread as a qualified lead and add a note"

Tag and tracker management

• "Tag this call as 'lead' and add a note"
• "Provision a new Google Ads call-extension tracker for Renaissance in area code 412" (requires confirm_billing=True, costs about $3/mo)

Installation

# Recommended: pipx for an isolated CLI install
pipx install callrail-mcp

# Or with pip
pip install callrail-mcp

To install from source (latest unreleased):

pipx install git+https://github.com/pghdma/callrail-mcp.git

Auth

Get an API key at Settings > API Keys in your CallRail account. You need Account Admin permission to create one.

Provide it one of two ways:

Option 1: environment variable (recommended for most setups)

export CALLRAIL_API_KEY="your_key_here"

Option 2: key file

mkdir -p ~/.config/callrail
echo "your_key_here" > ~/.config/callrail/api-key.txt
chmod 600 ~/.config/callrail/api-key.txt

Or override with CALLRAIL_API_KEY_FILE=/path/to/key.txt.

Optional: pin the account

Every tool resolves the account id with one GET /a.json per process and caches it. Set CALLRAIL_ACCOUNT_ID to skip that lookup, or to pin one account when the key can see several.

Configure your MCP client

Claude Code / Claude Desktop (~/.claude.json or claude_desktop_config.json)

{
  "mcpServers": {
    "callrail": {
      "command": "callrail-mcp",
      "env": {
        "CALLRAIL_API_KEY": "your_key_here"
      }
    }
  }
}

If you installed via pipx, callrail-mcp will be on your PATH automatically. Otherwise, point command at the full path to the executable.

Cursor and other clients

The server speaks standard MCP stdio. Any client that supports stdio MCP servers will work; just run callrail-mcp as the command.

Available tools

57 tools total, covering roughly 95% of CallRail's REST API v3 surface: read tools, write tools, tracker provisioning, agency aggregation, account management, notifications, integrations discovery, outbound calls, offline-lead backfill, leads and cross-channel timelines, SMS-thread lead management, and server-side analytics.

Read tools

Write tools

Tracker provisioning

Account management

Notifications and integrations

Outbound calling

Leads and server-side analytics

Agency aggregation

All tools accept account_id optionally; if omitted, the first accessible account is auto-resolved and cached for the process. Set CALLRAIL_ACCOUNT_ID to pin it and skip the lookup. Most tools accept company_id to filter to a single client.

A note on filtering calls

CallRail's GET /calls.json accepts answer_status (answered, missed, voicemail) but has no answered or source parameter. Earlier versions of this server forwarded both and CallRail silently ignored them, so results looked filtered but were not. I fixed that in v1.2.0:

• answer_status is the documented filter and is applied server-side.
• answered remains as a deprecated alias that translates to answer_status.
• source is applied client-side, and the response includes a source_filter block stating exactly what was matched. For source breakdowns, prefer call_stats(group_by="source").

Two more things CallRail does that bit me:

• Voicemail counts as answered on CallRail's side, because its system picked up. Since v1.3.0 call_summary, spam_detector, call_eligibility_check and bulk_update_calls treat a call as answered only when a person answered (answered and not voicemail) and report voicemail separately. answer_status=missed includes voicemail.
• No date window means CallRail's recent range, the previous 7 days including today, not all time. Windowed tools accept an IANA time_zone so day boundaries match the client's reporting zone, and usage_summary / compare_periods forward the account's zone automatically. Communication Records are retained for 25 months, so a window reaching past that returns an error asking for a narrower range.

Validation is strict throughout: phone-number format, area code (^\d{3}$), session pool size (CallRail's 4 to 50 range), name, whisper, and greeting length caps, and a 12-value source-type enum.

Rich field selection

The CallRail API returns a lean default payload. Ask for more fields on list_calls, get_call, or list_form_submissions via the fields parameter:

fields=company_name,source_name,keywords,landing_page_url,device_type,first_call,value,tags,note,gclid,fbclid,utm_source,utm_medium,utm_campaign,utm_content,utm_term,referrer_domain

Conversation-intelligence and attribution fields are also available by name: lead_status, sentiment, call_summary, call_highlights, keywords_spotted, transcription, conversational_transcript, speaker_percent, milestones, timeline_url, msclkid, prior_calls, total_calls, person_id. See the CallRail API docs for the full field catalog per resource.

Tool annotations

Every tool declares MCP tool annotations: readOnlyHint on the 33 read tools, destructiveHint on deletes and bulk writes, idempotentHint on updates, and openWorldHint on create_outbound_call because it reaches the phone network. Clients that honor annotations can auto-approve reads and ask before writes.

How this compares to CallRail's official MCP server

CallRail offers an official hosted MCP server (documented at apidocs.callrail.com) using OAuth 2.0, with roughly 30 tools and a server URL provided by your CallRail account team. It is a good option if you want a fully managed remote server.

I built this one for a different job:

Both speak the same underlying REST API v3. If you run an agency across many client accounts and want cost attribution and bulk workflows, this is the one I built for myself.

A community TypeScript server also exists, @kanopi/callrail-mcp on npm (34 tools, stdio, API key auth). What I put into this one instead is the agency tooling, the safety guards, and request contracts checked against CallRail's documentation and live responses (see the 1.3.0 entry in the CHANGELOG).

Out of scope (deliberately not implemented)

The following CallRail capabilities are not in this MCP, by design. PRs are welcome if you have an account that supports them, or open an issue and I will prioritize.

Blocked by CallRail account permissions (returns 403)

These endpoints exist but need account upgrades or permissions that my standard CallRail account does not have. I re-probed them live on 2026-10-05:

• Send SMS/MMS (POST /text-messages.json) needs A2P/TCR SMS registration. CallRail enforces TCPA-compliance keywords (STOP, CANCEL, UNSUBSCRIBE) on outbound text messages. MMS support was added by CallRail on 2026-05-05.
• Webhook integration create, update, delete (POST /integrations.json with type=Webhook) needs Integration-Admin permission.
• Outbound Caller IDs (/caller_ids.json CRUD) is documented by CallRail but returns 403 on a standard account.
• Message Flows (/message-flows.json CRUD, SMS auto-reply flows) is documented but returns 403.
• Integration Filters (/integration_triggers.json CRUD) is documented but returns 403.
• Form submission endpoints return 403 when no company on the account has Form Tracking enabled. That is account state rather than a permission: enable Form Tracking on one company and the endpoints open up. The tools say so in the error.

Not exposed by CallRail's REST API

These have no API equivalent and are managed exclusively via the CallRail web UI:

• Numbers: account-level number ownership, porting, and transfers.
• Call Flows: the IVR builder and call routing tree configuration.
• Custom Fields CRUD: custom data columns are readable as part of call and form responses, but the schema management endpoint is not exposed.
• Do Not Call list: DNC number management.

Available but not yet shipped

Readable on a standard account, but I have judged them low value so far: summary_emails CRUD, companies/bulk_update.json (external form capture only), and form_submissions/ignored_fields.json.

Examples

> List companies under our CallRail account.

(calls list_companies, returns clients with IDs and primary numbers)

> Pull today's calls for company COM019ab..., include source and keyword.

(calls list_calls with company_id, days=1, fields="source,keywords,landing_page_url")

> Why is my CallRail bill $174 this month? Break it down by client.

(calls usage_summary, returns per-company cost share sorted by biggest user)

> Why didn't this call show up as a conversion in Google Ads? CAL019dbf79...

(calls call_eligibility_check, returns gclid, duration, and answered checks
 plus a targeted reason such as "duration 58s under Google Ads minimum (60s)")

> Show me everything this lead has done with us.

(calls list_leads then get_lead_timeline, returns calls, forms, and texts in one timeline)

> Provision a new Google-Ads-call-extension tracker for Alan Construction in 412.

(calls create_tracker, refuses unless you also pass confirm_billing=True
 since it incurs a monthly charge)

Direct Python usage

The CallRailClient is also usable as a library:

from callrail_mcp.client import CallRailClient

cr = CallRailClient()  # picks up CALLRAIL_API_KEY
aid = cr.resolve_account_id()
for call in cr.paginate(f"a/{aid}/calls.json", {"per_page": 250}, items_key="calls"):
    print(call["id"], call.get("source"), call.get("customer_name"))

paginate() accepts an optional stats dict that reports pages_fetched, items_yielded, total_records, and truncated, so you can tell a capped result from a complete one.

Running the server directly

For debugging, or to verify your key works:

python -m callrail_mcp

The server speaks MCP stdio. It will wait for JSON-RPC messages on stdin. Press Ctrl-C to exit.

To smoke-test the API key without running the MCP loop:

python -c "from callrail_mcp.client import CallRailClient; c=CallRailClient(); print(c.get('a.json'))"

Rate limits

CallRail's documented default limits are 1,000 requests per hour and 10,000 per day for general API requests, with separate budgets for SMS sends (150 per hour, 1,000 per day) and outbound calls (100 per hour, 2,000 per day). The client retries 429 responses using the Retry-After header, and retries 5xx responses with exponential backoff for idempotent methods only (GET, PUT, DELETE, HEAD, OPTIONS). POST is never retried on 5xx, so a lost response cannot create duplicate trackers. For heavy pagination, prefer the built-in paginate() helper.

Development

git clone https://github.com/pghdma/callrail-mcp
cd callrail-mcp
python -m venv .venv && source .venv/bin/activate
pip install -e ".[dev]"
pytest

See DEVELOPMENT.md for release flow, API quirks worth knowing, and the conventions this codebase follows.

Contributing

Contributions welcome. See CONTRIBUTING.md for dev setup, test conventions, and release flow. Please file issues via GitHub Issues and follow the Code of Conduct.

Security

If you discover a security vulnerability, please report it privately per SECURITY.md instead of opening a public issue.

Author

Steve Japalucci, founder of Pittsburgh Digital Marketing Agency. I use this on my own client accounts every day, so bugs get noticed and fixed fast. Reach out at s@pghdma.com.

License

MIT. See LICENSE. Copyright (c) 2026 Steve Japalucci / Pittsburgh Digital Marketing Agency.

Disclaimer

This project is an independent open-source integration and is not affiliated with, endorsed by, or officially supported by CallRail. "CallRail" is a trademark of CallRail, Inc. All product names, logos, and brands are property of their respective owners.