Connect AI clients with the PRLane MCP server
Give Claude, ChatGPT, Codex, Cursor or any other MCP client read access to your pull request fleet, plus optional writes, with OAuth permissions you approve and can revoke.
Two kinds of connection
PRLane runs a remote MCP server over streamable HTTP. There is nothing to install: you paste a connection URL into your AI client and sign in to PRLane. Each URL is locked to what it covers, so a client can never switch itself to another workspace or organization.
| Workspace connection | Organization connection | |
|---|---|---|
| Covers | One workspace: its native projects, tasks and cached PR fleet | Selected repositories, or every repository linked to an active workspace in the org |
| Sign-in | OAuth, or a workspace token for clients without OAuth | OAuth only |
| Base scope | workspace:read | org:conflicts:read |
| Fleet tools | Always included | Only if you allow fleet:read on the consent screen |
| Best for | One team's PRs and tasks | Conflicts and PR status across many repositories |
Get your connection URL
Workspace connection
- Open the workspace, then Settings → MCP setup.
- Copy the Connection URL. It looks like
https://www.prlane.com/api/orgs/<org>/workspaces/<workspace>/mcp.
Organization connection
- Open Organization settings → MCP connections.
- Name the connection and choose Selected repositories (a fixed list of up to 100) or Entire organization (includes repositories added later).
- Create it and copy its URL. It looks like
https://www.prlane.com/api/orgs/<org>/mcp-conflicts/<connection-id>.
Creating a connection grants nothing by itself. Access starts when someone adds the URL to a client and approves PRLane's consent screen, and each approval is a separate session you can revoke.
Connect a client
The examples use a workspace URL. For an organization connection, paste its URL instead; everything else is the same. The settings page shows these snippets with your real URL filled in.
Claude (web, desktop and mobile)
- Open Settings → Connectors and choose Add custom connector.
- Paste the connection URL and leave Advanced settings empty.
- Select Connect, sign in to PRLane and review the consent screen.
Custom connectors are on Claude Pro, Max, Team and Enterprise. On Team and Enterprise an Owner adds the connector first, then each member connects.
Claude Code
claude mcp add --transport http prlane_workspace https://www.prlane.com/api/orgs/<org>/workspaces/<workspace>/mcpThen run /mcp in Claude Code and choose prlane_workspace to sign in.
ChatGPT
- Turn on Developer mode in Settings → Apps → Advanced settings.
- Choose Apps → Create, paste the connection URL and pick OAuth. Leave the client ID and secret empty.
- Scan tools, create the app, then sign in to PRLane and approve.
Developer mode is on ChatGPT Business, Enterprise and Edu on the web. Your workspace admin may need to allow it.
Codex
[mcp_servers.prlane_workspace]
url = "https://www.prlane.com/api/orgs/<org>/workspaces/<workspace>/mcp"Then run codex mcp login prlane_workspace to sign in.
Cursor
{
"mcpServers": {
"prlane_workspace": {
"url": "https://www.prlane.com/api/orgs/<org>/workspaces/<workspace>/mcp"
}
}
}Cursor opens the PRLane sign-in when you enable the server.
Other MCP clients
Any client that supports remote MCP servers over streamable HTTP with OAuth 2.1 (dynamic client registration and PKCE) can connect with the URL alone. See OAuth details below.
Workspace tokens (clients without OAuth)
On the workspace's MCP setup page, create one token per client (read-only, or read plus personal PR stars). It is shown once. Put it in the client's environment as PRLANE_MCP_TOKEN and send it as a bearer token. Never commit it.
claude mcp add --transport http prlane_workspace https://www.prlane.com/api/orgs/<org>/workspaces/<workspace>/mcp \
--header "Authorization: Bearer $PRLANE_MCP_TOKEN"[mcp_servers.prlane_workspace]
url = "https://www.prlane.com/api/orgs/<org>/workspaces/<workspace>/mcp"
bearer_token_env_var = "PRLANE_MCP_TOKEN"Scopes and permissions
Your client asks for scopes; PRLane's consent screen shows what each one allows before you approve. Read access is the default. Every write is opt-in twice: the client must request the scope and you must tick its box on the consent screen. A requested scope alone never grants a write.
| Scope | What it allows |
|---|---|
workspace:read | Read one workspace's projects, tasks and cached PR fleet. |
org:conflicts:read | Read cached merge conflicts for the repositories an organization connection covers. |
fleet:read | Organization connections only: add the Fleet read tools (dashboard, PR detail, reviews, labels, metrics, repair brief). |
fleet:star | Set your own PRLane star on a PR. It never stars anything on GitHub. |
fleet:refresh | Queue a refresh of one PR through PRLane's background worker, within its rate limits. |
fleet:labels | Create, rename and delete repository labels on GitHub, and add or remove labels on covered PRs. |
offline_access | Keep the connection alive: one-hour access tokens are renewed with rotating refresh tokens for up to 30 days. |
offline_access and will not fall back to a one-hour connection. If a client leaves it out, the consent screen blocks the connection: remove the connector and add it again so the client reads PRLane's current requirements. Renewal never extends the 30-day limit, and seven days without renewal ends access sooner.Tools
A client only sees the tools its grant allows. "Fleet read" means always on a workspace connection, and on an organization connection only when fleet:read was approved. Tools read PRLane's cache, so results can lag GitHub; each row carries timestamps such as updatedAt to check freshness. No tool merges a pull request.
| Tool | What it returns or does | Needs |
|---|---|---|
list_pull_requests | Cached open, closed or merged PRs for the workspace, with review, check and conflict state. | workspace:read |
list_repository_conflicts | Cached merge conflicts across the repositories the connection covers. | workspace:read or org:conflicts:read |
get_fleet_dashboard | Fleet rows, dashboard metrics and Kanban counts, with filters applied before paging. | Fleet read |
get_fleet_pull_request | One PR: summary and body, checks with status and links, or the estimated CI job cost ledger. | Fleet read |
list_fleet_reviews | Reviews, decisions, requested reviewers and review threads for one PR (a live GitHub read). | Fleet read |
get_fleet_repair_brief | A ready-to-use AI repair prompt for one open PR, built from its blockers and evidence. | Fleet read |
get_fleet_metrics | Timeframe totals, hourly or daily charts, agent and human counts, and a per-repository breakdown. | Fleet read |
get_fleet_release_comparison | How far main is ahead of the last production release for one repository, from the cached baseline (commit counts, not PR counts). | Fleet read |
list_fleet_repository_labels | Label definitions for a covered repository. | Fleet read |
set_pull_request_star | Star or unstar a PR for you only. | fleet:star |
request_fleet_pr_refresh / get_fleet_refresh_job | Queue a refresh of one PR and poll the job. Queued is not finished, and a refresh never re-runs CI. | fleet:refresh |
create_, update_, delete_fleet_repository_label / apply_, remove_fleet_pr_label | Manage repository labels and the labels on covered PRs. | fleet:labels |
OAuth details for client developers
PRLane follows the MCP authorization spec. An unauthenticated request to a connection URL answers 401 with a WWW-Authenticate header that names the resource metadata document and the scopes to request. From there a client discovers everything else.
| Endpoint | URL |
|---|---|
| Authorization server metadata | https://www.prlane.com/.well-known/oauth-authorization-server |
| Protected resource metadata | https://www.prlane.com/.well-known/oauth-protected-resource/<connection path> |
| Dynamic client registration | https://www.prlane.com/api/mcp/oauth/register |
| Token | https://www.prlane.com/api/mcp/oauth/token |
| Revocation | https://www.prlane.com/api/mcp/oauth/revoke |
- Grant types:
authorization_codeandrefresh_token. Response typecode, with PKCES256required. - Clients are public (
token_endpoint_auth_method: none) and register dynamically with one to five redirect URIs, which must be HTTPS or a loopback address. - The
resourceparameter is the connection URL, so a token is bound to one workspace or one organization connection. - Access tokens last one hour. Refresh tokens rotate on every use; the session ends after 30 days, or after 7 days without a refresh.
- Authorization responses include
iss. Tools also declare their required scopes in_meta.securitySchemes.
Revoke access and troubleshoot
- Revoke: organization sessions are listed under each connection in Organization settings → MCP connections; workspace tokens are on the workspace's MCP setup page. Revoking ends that session; other clients stay connected.
- Check renewal: open Session history and renewal diagnostics on an organization connection and look for Renewable and Refresh issued: Yes. That proves PRLane issued a refresh token, not that the client stored it, so confirm a renewal after the first hour before relying on it.
- Blocked on the consent screen: the client did not request
offline_access. Remove the connector and add it again, or ask the client's administrator to allow it. - A tool is missing: the grant doesn't include it. Fleet and write tools need their scope and their consent checkbox; reconnect and tick the permission you want.
- Data looks old: tools read PRLane's cache. Compare
updatedAt, or userequest_fleet_pr_refreshif you grantedfleet:refresh.