MCP Reference

MCP Resources & Prompts

Tools are the part of MCP everyone talks about, and they are only one of three things the Routebase server exposes. Alongside the 401 tools sit 12 resources — read-only data addressed by URI — and 13 prompts, guided workflows the server assembles from your live data.

The distinction matters in practice:

  • A tool is an action the model decides to take. It needs a scope, it can change something, and every call is checked server-side.
  • A resource is a document the client can read. It has a URI, it never changes anything, and it is meant to be attached to the conversation as context rather than "called".
  • A prompt is a workflow you start deliberately. The server loads the relevant state — your style guide, your open findings, your deprecation tracker — and hands the agent a task description grounded in it, so the first message is already about your API instead of a generic one.

Most MCP clients surface these differently: tools appear in the agent's toolbox, while resources and prompts are usually something you pick from a menu (in Claude, the connector's attachment and prompt pickers). If your client shows only tools, it does not support the other two — nothing is broken, and everything below has a tool equivalent.

All of them run inside the session context. Call set_context with your organization — and, for everything marked project below, a project — before reading a resource or starting a prompt. See Sessions in the MCP Quickstart.

Resources

Twelve URI templates, grouped by what they are anchored to. Everything here is read-only; there is no resource that writes.

Organization

URI Returns
routebase://orgs/{orgId}/style-guide Style guide rules and their effective severity levels.

Project

Needs a project context.

URI Returns
routebase://projects/{projectId}/environments All environments with their variables. Secret values are masked — see MCP Authentication.
routebase://projects/{projectId}/test-summary Aggregated test coverage and health across all suites in the project.
routebase://projects/{projectId}/mock-server Mock server configuration and URL. The access token is not part of it.
routebase://projects/{projectId}/monitoring Monitor status, 24 h uptime and p95, plus every incident still open or acknowledged. Request headers and bodies are deliberately omitted — use get_monitor for a monitor's full configuration, or list_incidents for an incident's event history.

Specification

Needs a project context.

URI Returns
routebase://specs/{specId} The complete spec exported as OpenAPI YAML.
routebase://specs/{specId}/endpoints Every endpoint with method, path, summary and parameters (JSON).
routebase://specs/{specId}/schemas Every schema defined in the spec (JSON).
routebase://specs/{specId}/changelog Structured changelog of the latest version: breaking, non-breaking and deprecated changes.
routebase://specs/{specId}/validation-report Style guide violations grouped by severity.
routebase://specs/{specId}/versions/{versionId} One specific version, exported as OpenAPI YAML.

Documentation

URI Returns
routebase://docs/{docId}/pages All documentation pages with their Markdown content, from the latest published version. Needs a project context.

The spec export resources are the ones worth reaching for first: routebase://specs/{specId} hands the agent the whole contract as OpenAPI in a single read, which is usually cheaper and more accurate than walking the endpoint tools one by one.

Prompts

Thirteen workflows. Each one loads live state before the agent starts, so the conversation opens with your data rather than a blank instruction.

Designing and reviewing

Prompt Arguments What it does
design_api description, style (REST default, or GraphQL) Designs a new API from a plain-English description — and loads your organization's style guide rules first, so what it produces is aimed at passing your own lint rules rather than generic conventions.
review_spec specId Reviews an existing spec for best practices, consistency and completeness, working from the spec details plus the current validation report. Project.
fix_violations specId Loads the current lint report and works through each violation with instructions to resolve it. Project.
generate_schema exampleJson, schemaName Turns an example JSON object into a proper schema definition. The one prompt that needs no context at all.

Building around the API

Prompt Arguments What it does
generate_tests specId, coverage (basic default, or thorough) Builds a test suite across the spec's endpoints. thorough adds edge cases and error scenarios; basic stays on the happy path. Project.
analyze_test_failures testRunId Loads a finished run's results and diagnoses each failure.
generate_mock_rules specId, includeErrors (default on) Creates mock responses for every endpoint, with 400/404/500 rules alongside the success cases unless you turn them off. Project.
document_api specId, audience (developer default, or manager) Writes guides, tutorials and examples pitched at the audience you name — code examples for developers, capability overviews for managers. Project.

Running the API

Prompt Arguments What it does
triage_security_findings severity (optional: info, low, medium, high, critical) Works through the project's open security findings, ordered by severity, guiding verification, remediation and the status decision on each. Project.
triage_incidents monitorId (optional) Works through the monitoring incidents that are still open or acknowledged: diagnosis, acknowledgement, resolution. Omit the monitor to triage all of them.
migrate_version specId, targetVersion Plans a new version with breaking-change analysis and a migration strategy, starting from the versions that already exist. Project.
run_deprecation — Walks a deprecation through its lifecycle — announce, notify consumers, advance, retire — loading everything the organization currently has deprecated and the rules it is checked against.
onboard_project projectId (optional, defaults to the session project) Produces a project overview for a new team member: specs, test health, mock server status, documentation.