Connecting to the Studio MCP
Source: studio_guide("mcp") (GET /api/studio/v1/guide/mcp) rendered at origin/platform (1afd83f), synced 2026-09-28. Do not edit this page here; change the source and run yarn sync:studio.
URLs
- Preview:
https://preview.bicycle.ai/mcp(every toolset exceptchat). Build and test here. - Prod:
https://app.bicycle.ai/mcp. - One toolset:
/mcp/<toolset>, for example/mcp/workflow.
The MCP is stateless: each request is its own connection. Long work returns an id at once; you poll it
(invocation_get, workflow_run_describe, agent_result, analysis_result).
Two ways to sign in
OAuth (Claude connector, claude mcp add without a header) | Platform API token | |
|---|---|---|
| How | the person consents in the browser | a person makes a token (role API) in Bicycle; --header "Authorization: Bearer $TOKEN" |
| Scopes | apps:*; functions:* only where the deployment offers them | the role's scopes, including functions:* |
| You see | dataapp_*, query_*, design_*, analysis_*, workflow_*, agent_*, notebook_*, invocation_* | all of those plus function_*, fn_*, wf_* |
No function_* tools? Call studio_whoami: it lists the scopes you hold and what to tell the person. A connection
keeps the scopes it was given until the person reconnects. tools/list shows only the tools your scopes allow.
Scopes
| Scope | Lets you |
|---|---|
apps:read | read apps, run queries, read workflows and runs, run Detect and Explain and registered agents |
apps:write | create apps and versions; create, change and run workflows |
apps:publish | publish apps and workflows (gated changes still need a person) |
functions:read / write / publish | read, author and test, publish functions (widening publishes need a person) |
functions:invoke | call fn_* and wf_* tools and read their runs |
Toolsets
| Prefix | For |
|---|---|
studio_ | this guide and studio_whoami; always on |
dataapp_ | apps: create, upload, publish, share, schedules, ask an app's chat |
query_ | the semantic catalog and cache-only semantic SQL |
design_ | compose an app from a spec by interview |
function_ | author, test and publish functions of every kind |
fn_<name>, wf_<slug> | call published functions and workflows exposed to MCP |
invocation_ | read any run: record, events, trace, lists |
workflow_ | build, try, publish and run workflows |
agent_ | run registered Bicycle agents such as cause (agent functions you build are function_*) |
analysis_ | Detect and Explain |
notebook_ | notebooks (not needed for data apps) |
Resources and prompts mirror the guides: studio://guide, prompt start-here, studio://skill/workflow,
studio://skill/ask-show-ship, studio://skill/semantic-query.
guide_version bb2461328de4