AI & MCP
Your AI, not ours
Fetch has an MCP server, API keys and OAuth, so Claude or any other agent can drive it. We don't sell AI: anything AI in Fetch runs on your own provider key.
The MCP server
https://fetch.nightroll.app/mcpStreamable HTTP, JSON responses. Every tool call is checked against the key's scopes and written to your organization's audit log.
With an API key
For Claude Code, scripts and any MCP client that sends headers. Create a key in Settings → API keys (scoped, with an expiry), then:
claude mcp add --transport http fetch https://fetch.nightroll.app/mcp --header "Authorization: Bearer $KEY"With OAuth
For claude.ai and Claude Desktop: add the URL above as a custom connector. You sign in, pick the organization and the permissions, and approve. Disconnect any time from Settings → API keys.
What your agent can do
- Import Postman collections, OpenAPI specs and cURL commands
- Create, edit and organize requests, folders and environments
- Send requests from Cloudflare and check the responses
- Run collections and set up monitors with assertions
- Publish mock servers and API docs
- Push and pull workspaces with GitHub
Tools
An agent sees only the tools its scopes allow: read, write or admin, never more than the person's role.
| Tool | What it does | Scope |
|---|---|---|
orgs_export | Export everything in an organization (workspaces, collections, environments, specs, history) as NDJSON: one entity per line. | admin |
workspaces_list | Workspaces you can see in an organization. | read |
workspaces_create | Create a workspace. | write |
workspaces_update | Rename a workspace or change its description or visibility. | write |
workspaces_delete | Move a workspace and everything in it to the trash (restorable for 30 days). | write |
items_list | List items (collections, folders, requests, environments, specs, mocks, monitors...) by workspace, kind or parent, or search them by name. | read |
items_get | One item with all its fields. | read |
items_tree | A collection or folder with all its folders, requests and examples, nested. | read |
items_create | Create an item: a collection, folder, request, example, environment, spec, package and so on. | write |
items_update | Change an item's name, place or fields. | write |
items_delete | Move an item (and what's inside it) to the trash, restorable for 30 days. | write |
items_restore | Take an item out of the trash. | write |
items_purge | Delete an item from the trash for good. | write |
items_duplicate | Copy an item and everything inside it. | write |
items_history | Version history of an item: every change, who made it and when. | read |
items_revert | Put an item back the way it was at a point in its history. | write |
search | Search everything in the organization by name, URL and description. | read |
catalog_list | The API catalog: every collection and spec with its docs, mocks and MCP servers. | read |
files_upload | Get a link to upload a file (request bodies, test data) straight to storage. | write |
files_confirm | Confirm an upload finished. | write |
files_url | A short-lived download link for a file. | read |
github_repos | Whether you've connected GitHub, and the repositories you can link (most recently pushed first). | read |
github_disconnect | Forget your GitHub connection for this organization. | write |
git_status | A workspace's GitHub link and what differs: files changed in Fetch (push sends them), on GitHub (pull brings them) or both (conflicts). Collections and environments are kept as Fetch YAML files. | read |
git_link | Link a workspace to a folder of a GitHub repository and branch, or unlink it (empty `repo`). | write |
git_pull | Bring GitHub's changes into the workspace (and settle conflicts with `resolve`). | write |
git_push | Commit the workspace's changes to GitHub (pull first when GitHub has changes). | write |
git_diff | A changed file in the workspace and on GitHub. | read |
git_log | Recent commits to the linked folder. | read |
git_branches | The linked repository's branches. | read |
requests_send | Send a request through the cloud executor (public internet only) and get the response, test results and console output. For localhost or private networks, use the Fetch Agent instead. | write |
runs_start | Run a collection or folder (optionally many iterations with CSV/JSON data) in the cloud. Returns the run, which progresses in the background. | write |
runs_get | A run's status and results. | read |
runs_list | Recent runs (collection runs, monitors, CI, contract tests). | read |
runs_cancel | Stop a run. | write |
runs_report | A run's report as JUnit XML, JSON, HTML or CTRF. | read |
history_list | Your requests sent from the cloud in a workspace, newest first, kept 30 days: method, URL, status, time, size, tests and error. history.get adds the request and response. Browser and agent sends stay on their device. | read |
history_get | One cloud send with its request and response (the first 256 KB of the body, kept 30 days): a history entry, or a request of a cloud collection run or monitor check (the run's id and the request's `step` in its results). | read |
history_clear | Clear a workspace's request history. | write |
monitors_create | Create a monitor: run a collection on a schedule and alert on failures. | write |
monitors_run | Run a monitor now. | write |
import | Import a file: cURL, Postman (v2.0, v2.1, v3 YAML), OpenAPI 3.x, Swagger 2.0, GraphQL or Fetch YAML. The format is detected. | write |
export | Export a collection, folder, request or environment as Fetch YAML, Postman v2.1, OpenAPI 3.1 or cURL. | read |
codegen | Generate code that sends a request, in any of 20+ languages and libraries. | read |
codegen_targets | Languages and libraries codegen can write. | read |
curl_parse | Turn a cURL command into a request. | read |
specs_to_collection | Generate a collection from a spec (or update the linked one). | write |
collections_to_spec | Generate an OpenAPI 3.1 spec from a collection. | write |
docs_publish | Publish (or unpublish) a collection's documentation, with an llms.txt for AI tools. | write |
mocks_create | Create a mock server from a collection's saved examples. | write |
hosted_url | The public URL of a mock, catcher, hosted MCP server or published docs. | read |
whoami | Who the key or token belongs to: person, organization, role and scopes. | read |
list_members | The organization's members and their roles. | read |
invite_members | Invite people by email, with a role. | admin |
list_invites | Open invites. | admin |
revoke_invite | Revoke an open invite. | admin |
get_usage | Usage and cost for a month, per meter. | admin |
list_audit_events | The audit log, filtered by person, action and dates. | admin |
Bring your own key
Fetch's in-app assistant runs in your browser on your own Anthropic, OpenAI or OpenRouter key (or any OpenAI-compatible endpoint); the key stays on your device. Your key stays in your browser: it never reaches our servers, and your provider bills you directly.
We don't sell AI
No AI credits, no markup on tokens, no training on your data. You choose the model and the provider, and you pay them directly.
Connect in a minute
Step by step, with example prompts.