Routebase

5.0 Agents

Make every API agent-ready

Your workspace ships a Model Context Protocol server. Describe the API you want — your agent designs the spec, writes the tests, and runs them, under your team's permissions.

Start free

14-day Pro trial — no credit card required.

Describe the API. Get the spec — and the tests.

Two shipped prompts carry the whole job: design_api builds the contract, generate_tests turns it into a suite that runs.

• design_api walks it call by call: create_spec, create_schema, create_endpoint, add_response, validate_spec
• Your organization's style guide is loaded as context first, so the draft arrives already linted
• generate_tests covers happy path, 400, 401, 403, 404, response schema, latency, and edge cases
• run_test_suite executes the suite and reports which assertion failed, on which case
• The agent works on a draft version — nothing reaches consumers until someone publishes

Try it with your agent — free

What comes back is your workspace, not a transcript

No wall of YAML to paste. Every call creates a real object in Routebase — one your team can open, diff, and review.

• Endpoints, schemas, and folders land in the visual editor
• Agent work can run on a spec branch and arrive as a merge request
• validate_spec has a dry-run mode: same errors and counts, nothing written
• publish_version is the one step that freezes an immutable OpenAPI snapshot

The rest of the lifecycle, same session

Design is where it starts. Mocks, monitors, security triage, and docs are the same connection — 400+ tools across 31 toolsets.

• generate_rules_from_spec turns the new endpoints into a running mock
• generate_monitors_from_spec puts the same contract under monitoring
• 87 testing tools — suites, scenarios, runs, results, schedules
• Monitoring, Security, and Mock Server are fully agent-reachable, enforced by a CI coverage ratchet

Context and whole workflows, prepackaged

Beyond tools: 12 read-only resources hand agents context without spending a call, and 13 prompts package multi-step work into one invocation.

• Resources: spec as OpenAPI YAML, changelog, validation report, test summary, style guide
• Design prompts: design_api, review_spec, fix_violations, generate_schema
• Operations prompts: analyze_test_failures, triage_security_findings, triage_incidents, run_deprecation
• Structured logs and usage metrics per session

A lean session by default

400+ tool definitions would fill a context window before the agent does anything. A default session sees 48.

• Five core toolsets: context, navigation, projects, specs, endpoints
• About 6k tokens of tool definitions instead of 55k — an 88% cut
• enable_toolset adds what the task needs, mid-session
• A scoped API key or an OAuth sign-in skips the staging entirely
• Hidden is never disabled — a tool stays callable by name

Agents under governance

An agent is a user, not a backdoor. Every call runs through the same RBAC your team already has — once at discovery, again at execution.

• Agents only see tools their permissions allow
• OAuth 2.1 sign-in or a scoped API key
• Billing and org access stay read-only: an agent explains a limit, it never grants a role or spends money
• 200 requests per minute per user, separate from the REST API
• EU or US region — the same choice as the app

Connect in a minute — or with nothing installed

Claude adds the server as a custom connector and signs you in. Every other client is one command.

• Remote connector at mcp.routebase.dev for Claude web, desktop, and mobile — OAuth, no API key, nothing on disk
• Claude Code, Cursor, and VS Code talk to the hosted endpoint over streamable HTTP
• npx routebase-mcp init writes the config for your IDE
• stdio bridge with self-contained binaries for macOS, Windows, and Linux

Practicing what we preach: this site ships llms.txt

Frequently asked questions

Do I need an API key to try this?

Not with Claude. Add https://mcp.routebase.dev as a custom connector in Claude web, desktop, or mobile and sign in with your Routebase account — the agent gets exactly your permissions, and nothing is stored on disk. Claude Code, Cursor, and VS Code use an API key with the scopes you choose.

Will 400+ tools blow up my context window?

No. A default session lists 48 core tools — roughly 6k tokens of definitions instead of 55k — and the agent enables further toolsets when a task needs them. A scoped API key or an OAuth sign-in skips that staging and lists everything you are entitled to right away.

Can an agent break something?

It can do exactly what the person or key behind it can do, and no more: permissions are checked when tools are listed and again when one is called. Billing and organization administration are read-only over MCP, specs stay drafts until someone publishes a version, and every step is an object you can review or revert.

What does the agent actually produce?

Real objects in your workspace — spec versions, endpoints, schemas, mock rules, test suites and their runs — not a chat transcript to copy out. You review them in the editor, optionally as a merge request on a spec branch, and export the spec as OpenAPI whenever you want.

Where does the data live?

In the region your organization chose, EU or US — the MCP endpoint follows the same choice as the app. Connectors select it with a query parameter, the CLI with an environment variable.

One living spec

The rest of the lifecycle

Change the spec — everything on this page updates with it.

1.0

Design

Design the contract first

2.0

Mock

Start frontend work on day one

3.0

Test

Prove reality against the spec

4.0

Document

Publish docs that never drift

Ready to ship on it?

Routebase is live. Design your API once — docs, mocks, tests, and monitoring all follow from the same source.

Start free