Publish an MCP endpoint
Create an endpoint, choose tools and a permission preset, issue access tokens and understand rate limits.
An MCP endpoint is the URL an AI client connects to. It combines tools from one or more integrations, each called through a connection you choose, and applies a permission preset plus any policies you define.
Endpoint URL
https://mcp.pipmcp.com/<workspace>/<endpoint>
<workspace> is your workspace slug and <endpoint> is the endpoint slug, both lowercase letters, digits and dashes. The endpoint page shows the full URL with a copy button.
Create an endpoint
- Open MCP Endpoints and choose Create endpoint.
- Name it. The slug is derived from the name and can be changed.
- Add one or more integrations and pick the connection each one should use.
- Choose a safety preset (see below).
- Save, then create an access token.
Only published tool versions are served. If you edit a tool or import a new version, agents keep getting the previously published version until you publish again (see Versioning).
Choose which tools are exposed
By default an endpoint exposes every enabled, published tool of its integrations. Switch the tool selection to Selected to expose only the tools you tick. Tools disabled on the integration are never exposed.
Keep endpoints focused. Agents choose better from a short list; if an endpoint exposes more than about 80 tools, the dashboard warns you. Each plan also caps the number of tools one endpoint can expose.
A master key is one URL for the whole workspace. It lists only tools marked Master key on each API. An API with none of those contributes nothing. e-conomic's master-key list is a short set of reads; its own endpoint still lists the full API. Each link can be limited to specific APIs. A link with no limit sees every master-key tool, including APIs you connect later. A limited link does not.
Safety presets
| Preset | Read-only (0) | Low-impact write (1) | External effect (2) | Destructive (3) | Financial (4) |
|---|---|---|---|---|---|
| Read only | Runs | Blocked | Blocked | Blocked | Blocked |
| Safe writes (default) | Runs | Runs | Runs | Confirmation | Confirmation |
| No destructive actions | Runs | Runs | Runs | Blocked | Blocked |
| Confirm external effects | Runs | Runs | Confirmation | Confirmation | Confirmation |
| Custom | Runs | Runs | Runs | Confirmation | Confirmation |
With Custom, your policies decide first and the table above is the fallback. Policies always take precedence over presets. See Permissions and risk.
Access tokens
Agents authenticate with an endpoint access token sent as a bearer token:
Authorization: Bearer pmcp_...
- Tokens start with
pmcp_and are shown once, when you create them. PipMCP stores only a hash. - A token belongs to exactly one endpoint and cannot be used with another.
- Scopes:
tools:listandtools:call. A token with onlytools:listcan discover tools but not run them. - Tokens can have an expiry date and can be revoked at any time on the endpoint page. Revocation takes effect immediately.
- A token can carry an access role. The role limits tools on integrations you have restricted from the integration's Tools tab. A token with no role can use every tool this endpoint exposes. See Permissions and risk.
Create one token per person, with a role when people in the office should have different tools. The executions log then shows who called what, and you can revoke one person without touching the others.
Approval mode
When a call needs confirmation and the AI client cannot ask the user inline, the endpoint's approval mode decides what happens:
- Dashboard (default): the call is recorded as a pending approval. Someone approves it under Executions, and the agent retries the same call.
- None: the call is refused with
CONFIRMATION_REQUIREDand no approval is created.
See Permissions and risk.
Rate limits
MCP requests are limited per minute. The defaults are:
| Scope | Requests per minute |
|---|---|
| Per access token | 120 |
| Per endpoint | 600 |
| Per workspace | 1,200 |
When a limit is hit the response is HTTP 429 with a Retry-After header.
Disable or delete
Set an endpoint to Disabled to stop serving it immediately without losing its configuration; clients receive HTTP 403. Deleting an endpoint makes its URL and all of its tokens stop working.