Integrations · MCP
MCP server.
Connect Claude Desktop, Cursor, or any MCP-aware AI client directly to your TraceWorks workspace. No copy/paste, no sidecar exports — your assistant queries the same data you see in the web app, with the same scope and audit trail.
What is MCP?
The Model Context Protocol is an open standard for connecting AI assistants to external systems. Your client (Claude Desktop, Cursor, etc.) speaks MCP; TraceWorks runs an MCP server at /mcp. The two negotiate, you sign in with your TraceWorks credentials, and from that point the assistant can call tools that read or write your workspace.
What you can do
A few examples of prompts that work end-to-end once you’re connected:
“Show me every open risk on Project Atlas that doesn't have an owner assigned.”
listRisks + searchGlobal
“Draft a requirement based on this paragraph from the customer email, then create it under Project Atlas as DRAFT.”
createRequirement (write)
“What does the uploaded RFP say about thermal margins? Cite the chunks you used.”
queryProjectContext (RAG)
“Summarize blockers for the next sprint — which tasks are blocked, by what, and which requirements they're tied to.”
listTasks + getRequirement
Tool reference
Fifty-two tools today — 25 read and 27 write — covering every artifact module, the trace graph, coverage gaps, and RAG retrieval. Read tools accept any valid token; write tools require the mcp:write scope, granted during the OAuth consent step when the client connects. Writes inherit the same service-layer gates as the web app — edits to approved requirements route through a change request, and every write records an ActivityLog entry.
| Tool | Scope | Purpose |
|---|---|---|
| listProjects | read | Projects in your org with rollup counts |
| getProject | read | Project detail + recent activity |
| listRequirements | read | Requirements in a project, priority-ordered |
| getRequirement | read | One requirement with linked specifications |
| listSpecifications | read | Specifications in a project |
| getSpecification | read | One specification with sections and links |
| listTasks | read | Tasks with assignees and blocker links |
| listRisks | read | Risks with owner, impact, mitigation plan |
| searchGlobal | read | Title search across 7 record types |
| queryProjectContext | read | RAG retrieval over indexed source documents |
Connect Claude Desktop
Add this to your Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json on macOS, equivalent path on Windows):
{
"mcpServers": {
"traceworks": {
"type": "http",
"url": "https://www.traceworks.app/mcp"
}
}
}Restart Claude Desktop. On the first tool call, a browser window opens to the TraceWorks sign-in page. The authorized session maps directly to your workspace user.
Other clients
Any client that speaks MCP over HTTP works. Point it at https://www.traceworks.app/mcp and follow its OAuth setup flow — Dynamic Client Registration means you don’t need to pre-register your client with us.
Security
Every request is gated by an OAuth 2.1 bearer token issued by TraceWorks itself (PKCE-only clients, validated server-side on every call), tools delegate to the same service layer the web app uses so organization scoping is enforced uniformly, and every write tool records an ActivityLog entry. More detail on the security page.
Want a tool that isn’t listed?
Each tool is roughly 20-40 lines and wraps an existing service function. Tell us the workflow you want to automate and we’ll wire it up.
Request a tool