WordPress Trac MCP Server
OfficialThis read-only MCP server lets you query public WordPress Trac data without credentials.
Search Trac tickets by keyword or filter expression, with optional status/component filters and a result limit.
Fetch a specific ticket's description, metadata, comments, and discussion.
Fetch a changeset/commit by SVN revision, optionally with a truncated diff.
Read recent Trac timeline activity for a configurable number of days and events.
List Trac metadata: components, milestones, priorities, or severities.
Connect to a specific Trac instance; the endpoint determines which Trac the tools query.
The provided schema is narrower than the README: no ticket pagination, no timeline date/author filters, limited metadata types, and lower caps.
Enables deployment of the MCP server on Cloudflare Workers for global edge distribution of the WordPress Trac data service.
Supports ChatGPT's Deep Research feature with a simplified interface for searching WordPress Trac data and fetching detailed information about tickets and changesets.
Transforms WordPress.org Trac into an AI-accessible knowledge base, providing comprehensive access to WordPress tickets, code changes, and development activity.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@WordPress Trac MCP Server search for recent open tickets about the block editor"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
WordPress Trac MCP server
A read-only Model Context Protocol server for the WordPress.org Trac instances, starting with WordPress Core Trac. It runs as a Cloudflare Worker and uses Trac's public HTML, CSV, RSS, and diff endpoints.
Live servers:
Production:
Standard MCP:
https://wordpress-trac-mcp-server-prod.a8c-aiops.workers.dev/mcpSearch/fetch compatibility:
https://wordpress-trac-mcp-server-prod.a8c-aiops.workers.dev/mcp/chatgptHealth check:
https://wordpress-trac-mcp-server-prod.a8c-aiops.workers.dev/health
Staging:
Standard MCP:
https://mcp-server-wporg-trac-staging.a8c-aiops.workers.dev/mcpSearch/fetch compatibility:
https://mcp-server-wporg-trac-staging.a8c-aiops.workers.dev/mcp/chatgptHealth check:
https://mcp-server-wporg-trac-staging.a8c-aiops.workers.dev/health
The former staging deployment at https://mcp-server-wporg-trac-staging.a8cai.workers.dev is
deprecated and runs older code. Its a8cai.workers.dev subdomain differs from the active staging
deployment's a8c-aiops.workers.dev subdomain.
Set a user agent
Cloudflare runs a Browser Integrity Check
on every workers.dev hostname, and it cannot be turned off. It rejects some default client user
agents, such as Python's Python-urllib/3.x, with HTTP 403 and Cloudflare error 1010. The request
never reaches the server, and retrying does not help.
Send a User-Agent header that names your client. It does not need to look like a browser:
request = urllib.request.Request(url, data=body, headers={
"Content-Type": "application/json",
"Accept": "application/json, text/event-stream",
"User-Agent": "my-tool/1.0",
})The default user agents of requests, httpx, aiohttp, and curl already pass.
Trac instances
Each Trac instance has its own endpoint. /mcp and /mcp/chatgpt serve Core.
Trac | Standard MCP | Search/fetch compatibility |
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
|
The table is a discovery aid rather than an allowlist. Any <slug>.trac.wordpress.org resolves at
/mcp/<slug>, so a Trac added later needs no change here. An instance is bound to the connection
rather than chosen per tool call, so a client cannot read the wrong Trac by mistake.
Instances configure different fields. Themes has no components, and only some instances have
severities. getTracInfo reports a field the instance does not configure as unavailable instead of
failing. Ticket fields behave the same way: focuses exists only on Core and comes back empty
elsewhere.
Filtering is stricter, because Trac answers a filter on a field it does not configure with the
unfiltered result set rather than an error, and that reads as a real match count. searchTickets
rejects such a filter and names the fields the instance does have. This covers both the separate
arguments and the expressions inside query.
Connect to one instance per client entry. Use several entries to read several Tracs.
Related MCP server: WordPress MCP
Tools
The standard /mcp endpoint provides:
Tool | Purpose |
| Search by keywords, ticket number, or structured filters |
| Read a ticket, its attachments, changesets, human discussion, and linked pull requests. |
| Read a changeset and an optional truncated diff |
| Read Trac activity for recent days or a historical date range, with author filtering and day-granular coverage |
| List components, milestones, priorities, severities, types, or statuses |
getTicket leaves bot comments, cc-only changes, and entries with neither a change nor text out of
comments and lists each one under omittedComments with its ID, author, and reason (bot, cc,
or empty), so a gap in the comment numbering is explained rather than mistaken for truncation.
Every other field change a person makes is reported in changes with its values, including keyword
edits and description edits with their diff link.
getChangeset expects the numeric revision argument, not rev:
{
"revision": 58504,
"includeDiff": false
}The /mcp/chatgpt compatibility endpoint provides search and fetch. Use a bare number for a
ticket and an r prefix for a changeset: 65739 and r58504.
No tool takes a Trac instance argument. The endpoint you connect to decides which Trac the tools read.
Ticket and changeset text is plain text with one exception: links are kept as <a href="...">
with an absolute URL, because a comment that points at a pull request or another ticket loses its
point without one. Relative Trac links resolve against the instance you connected to.
Search filters
searchTickets accepts plain keywords, ticket numbers, or filter expressions joined with &.
Plain keywords match the ticket summary only; use description~=text to search ticket bodies.
{
"query": "milestone=6.9&status=closed&resolution=fixed",
"limit": 50,
"page": 2
}Expressions can name summary, description, owner, reporter, type, status, priority,
milestone, component, version, severity, resolution, keywords, cc, or focuses, and
take four operators: = exact, ~= contains, != not equal, and !~= does not contain. Repeat
a field to OR its values (status=new&status=assigned), with the same operator each time. Add order=<column> and desc=1 to
sort, for example component=Editor&status!=closed&order=changetime&desc=1. Sortable columns are
the fields above plus time and changetime. Field values differ by instance: getTracInfo
lists the complete components, milestones, priorities, severities, types, and statuses each one configures, in Trac's order. Milestones include open and closed groups as one flat list.
It also accepts status, component, milestone, and resolution as separate arguments. Each
is an exact match on one value and overrides the same field in query; the expression form is
the one to use for substring, OR, or negation. Results include pagination metadata.
Timeline ranges, authors, and coverage
getTimeline reads the last days days (default 7, max 30) or an explicit from/to date range. Historical dates start at 2005-01-01, the start of the verified WordPress Core Trac timeline (younger instances have no events before their own first day), and may span at most 90 days per request because the upstream timeline caps its lookback. Dates are inclusive UTC calendar days: days counts whole days ending today, from on its own ends at today, and to on its own covers the seven days ending at to. A to in the future is rejected, and days cannot be combined with from/to:
{
"from": "2005-01-01",
"to": "2005-01-31",
"author": "saxmatt",
"limit": 20
}author takes one Trac username or a list of up to ten. The server filters by author before the
event limit applies, so a contributor's events stay complete even inside a busy window.
Calls that use only days and limit keep the original recent-activity contract: limit is the maximum number of upstream events, and the response contains results, totalEvents, daysBack, and timelineUrl.
For a date-range or author-filtered call, limit (1 to 100, default 20) is advisory. A response covers whole calendar days: results are rounded down to a day boundary, and the newest complete day of the window comes back in full even when it holds more events than limit.
Each coverage response lists events newest first in results and reports requested, the window that was
asked for, and covered, the part of it this response covers completely. complete says whether
the response covered all of it. When it is false, continueWith is a ready-made arguments object
for the remainder: it preserves the effective limit and the accepted author input when present,
so send it back to getTimeline unchanged and repeat until a response reports complete as true,
or until a response arrives without a continueWith. A continuation window always ends on a day already past, so walking one cannot produce gaps or duplicates. note states the coverage in plain language, and coverage that includes today is accurate as of the request. If one day fills the 500-event upstream fetch, that incomplete day is not included in covered: covered is null, complete is false, and terminalTruncation states that the tool cannot continue within that day.
docs/timeline-pagination.md records why the timeline reports day coverage instead of page numbers.
Tool errors
A failed tool call returns an MCP result with isError: true. Its JSON payload carries a
machine-readable code alongside the human-readable error message. not_found errors also name
the resource and id that were requested:
{
"code": "not_found",
"error": "Ticket 99999999 not found",
"resource": "ticket",
"id": 99999999
}
| Meaning |
| The requested ticket or changeset does not exist |
| An argument passed schema validation but cannot be used, such as an unsupported search filter field |
| Trac throttled the request and bounded retries did not clear it |
| Trac or a supporting service failed or returned unexpected content |
Codes are stable API surface: branch on code, never on error wording. An existing code keeps
its meaning and is only removed or renamed with a major version bump, while messages can change
freely. New codes may be added over time, so treat an unrecognized code as upstream_error.
Arguments that fail the advertised input schema also return isError: true, but as the MCP SDK's
plain-text message beginning Input validation error: rather than a JSON payload, so a model can
read it and retry. An unknown tool name is a JSON-RPC -32602 error.
Connect
Remote-capable MCP clients can connect directly to the standard endpoint. Clients that need a local
bridge can use mcp-remote:
{
"mcpServers": {
"wordpress-trac": {
"command": "npx",
"args": [
"mcp-remote",
"https://wordpress-trac-mcp-server-prod.a8c-aiops.workers.dev/mcp"
]
},
"wordpress-meta-trac": {
"command": "npx",
"args": [
"mcp-remote",
"https://wordpress-trac-mcp-server-prod.a8c-aiops.workers.dev/mcp/meta"
]
}
}
}For ChatGPT, add the compatibility endpoint as a custom app. See OpenAI's current MCP help because product labels and setup steps change.
After changing a configured server URL, reconnect the MCP server or restart the client once. Future deployments to the same URL do not require a client configuration change.
Develop
Requirements: Node.js 22 or later and pnpm 10.
pnpm install
pnpm devOpen http://localhost:8787/ to view the local landing page. See
docs/local-development.md for the full browser-preview workflow and
troubleshooting.
Run the complete local quality gate:
pnpm checkThis runs TypeScript, Biome, Vitest, and a Cloudflare Worker dry-run build. See docs/testing.md for manual protocol and live-data checks.
Deployment requires a configured Cloudflare account:
# Staging
pnpm run deploy
# Production
pnpm run deploy:productionDesign and safety
The server is read-only and has no Trac credentials.
The MCP protocol layer is the official TypeScript SDK. Each endpoint serves stateless 2026-07-28 clients and handshake-era clients (2024-10-07 through 2025-11-25), answering the latter with plain JSON. The advertised input schemas are generated from the same Zod schemas that validate calls.
Tool inputs receive runtime validation before any upstream request.
Upstream requests stay on
*.trac.wordpress.organd the official linked-PR endpoint onapi.wordpress.org. The instance slug comes from the URL path, is validated against a strict pattern before it reaches a request, and every request is checked against the resolved origin.Upstream redirects are never followed.
*.trac.wordpress.orghas wildcard DNS and redirects unknown subdomains to Core, so following one would answer for one instance with another's data. A redirect that leaves the instance origin is reported as an unknown instance; one that stays on it is reported as an upstream failure.Transient transport failures, rate limits, server errors, and Trac bot challenges receive bounded retries. Permanent 403 and 404 responses return immediately.
Responses are parsed from public Trac pages and machine-readable formats.
The Worker keeps no ticket cache or durable state.
Contribute
Keep tool schemas, runtime validation, tests, and documentation aligned. Run pnpm check before
opening a pull request.
License
GPL-2.0-or-later.
Available Tools
5 toolsgetChangesetC
Get information about a specific WordPress code changeset/commit including commit message, author, and diff.
Inputs: revision (required), includeDiff, diffLimit.
getTicketC
Get detailed information about a specific WordPress Trac ticket including description, comments, and metadata.
Inputs: id (required), includeComments, commentLimit.
getTimelineB
Get recent activity from WordPress Trac timeline including recent tickets, commits, and other events.
Inputs: days, limit.
getTracInfoA
Get WordPress Trac metadata like components, milestones, priorities, and severities.
Inputs: type (required).
searchTicketsB
Search for WordPress Trac tickets by keyword or filter expression. Returns ticket summaries with basic info.
Inputs: query (required), limit, status, component.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 5 tools
Each tool has a clearly distinct purpose with no overlap: getChangeset retrieves commit details, getTicket fetches ticket information, getTimeline shows recent activity, getTracInfo provides metadata, and searchTickets searches for tickets. The descriptions clearly differentiate these functions, making misselection unlikely.
All tool names follow a consistent verb_noun pattern using camelCase (e.g., getChangeset, getTicket, getTimeline, getTracInfo, searchTickets). The naming is predictable and readable throughout the set, with no deviations in style or convention.
With 5 tools, this server is well-scoped for its purpose of interacting with WordPress Trac. Each tool serves a specific, non-redundant function, and the count is appropriate for covering key operations like retrieving changesets, tickets, activity, metadata, and searching without being overwhelming.
The tool set provides strong coverage for querying and retrieving information from WordPress Trac, including tickets, changesets, activity, metadata, and search. A minor gap exists in the lack of tools for creating or updating tickets or changesets, but this is reasonable for a read-only or query-focused server, and agents can still perform core workflows effectively.
Maintenance
Related MCP Connectors
A comprehensive Model Context Protocol (MCP) server that enables AI assistants to interact with yo…
A Model Context Protocol server for Wix AI tools
Enable secure connectivity between Sentry issues and debugging data, and LLM clients, using a Model Context Protocol (MCP) server.
Hosted WordPress MCP server for Claude, ChatGPT and Cursor, with each action switched per account.
Related MCP Servers
- AlicenseNot gradedqualityDmaintenanceA Model Context Protocol server that enables AI assistants like Claude to interact with GitHub repositories, issues, and pull requests.23 npmMIT

WordPress MCPofficial
AlicenseBqualityFmaintenanceA WordPress plugin that implements the Model Context Protocol to enable AI models and applications to interact with WordPress sites in a structured and secure way.548 npm946ISC- AlicenseNot gradedqualityDmaintenanceA Model Context Protocol server that enables AI assistants to manage WordPress sites and create content with AI-generated featured images.18 npm2MIT
- AlicenseNot gradedqualityDmaintenanceA Model Context Protocol server that turns WordPress sites into AI-operable surfaces by exposing CRUD primitives for posts, users, comments, and settings. It supports both local STDIO and remote HTTP transports, allowing AI assistants to manage content and site configuration directly.94 npm10MIT