Muse readiness SDK

Use gatherMuseReadinessEvidence and gradeMuseReadiness to check your MCP server against Meta’s Muse connector guidelines before submitting. The two functions are split so you can gather evidence on one machine and grade on another — or supply your own pre-fetched tool listing when the server requires credentials.

Muse readiness is a local preflight, not a submission guarantee. Meta reviews every connector by hand — a risk assessment, a per-tool review, and end-to-end QA — so a ready result means the stated requirements are met, not that approval is certain.

Import

import {
  gatherMuseReadinessEvidence,
  gradeMuseReadiness,
  formatMuseClassificationSheet,
} from "@mcpjam/sdk";

gatherMuseReadinessEvidence is a Node-only export (it dials the network). gradeMuseReadiness and formatMuseClassificationSheet are pure functions and are also available from @mcpjam/sdk/browser.

Quick example

import {
  gatherMuseReadinessEvidence,
  gradeMuseReadiness,
  formatMuseClassificationSheet,
} from "@mcpjam/sdk";

const evidence = await gatherMuseReadinessEvidence({
  enteredUrl: "https://your-server.com/mcp",
  fetchFn: fetch,
});

const result = gradeMuseReadiness(evidence);

console.log(result.technicalStatus); // "ready" | "not-ready" | "incomplete"
console.log(result.status);          // submission-ready rollup
console.log(result.summary);

// Print the §5.6 classification table
console.log(formatMuseClassificationSheet(result.classificationSheet));

gatherMuseReadinessEvidence()

Dials the server and collects everything gradeMuseReadiness needs. Returns a plain MuseReadinessInput object that survives JSON.stringify.

gatherMuseReadinessEvidence(
  options: GatherMuseReadinessEvidenceOptions
): Promise<MuseReadinessInput>

GatherMuseReadinessEvidenceOptions

Property	Type	Required	Description
`enteredUrl`	`string`	Yes	The connector URL exactly as the user entered it. Never canonicalized.
`fetchFn`	`typeof fetch`	No	The transport. Without it the gatherer dials nothing and every wire lane reports its gap — the honest outcome for a run assembled from supplied evidence alone.
`signal`	`AbortSignal`	No	Composed into every request so a cancelled run stops the one in flight.
`capabilities`	`MuseRunnerCapability[]`	No	Runner capabilities to declare (e.g. `["dns"]`).
`tools`	`DirectoryToolEvidence[]`	No	A tool listing the caller already holds — typically one read with the submitter’s credentials, since an OAuth connector refuses an anonymous `tools/list`. Supplying it skips the dial.
`submissionProfile`	`unknown`	No	The §5 submission profile. Validated during grading; issues become findings.
`evidenceSources`	`string[]`	No	Suite results consumed as evidence, named for the report.
`timeoutMs`	`number`	No	Per-request timeout in ms.
`maxRedirects`	`number`	No	Maximum redirects to follow.
`headers`	`Record<string, string>`	No	Extra HTTP headers.

Supplying a pre-fetched tool listing

When the server requires credentials, pass the listing you already hold:

const tools = await myAuthenticatedClient.listTools();

const evidence = await gatherMuseReadinessEvidence({
  enteredUrl: "https://your-server.com/mcp",
  tools: tools.tools, // skips the dial
  submissionProfile: myProfile,
});

An explicit empty array (tools: []) tells the grader the server advertises no tools — it is not the same as omitting tools.

gradeMuseReadiness()

Grades gathered evidence. Pure — no network, no clock, no randomness. Same evidence in, same result out.

gradeMuseReadiness(input: MuseReadinessInput): MuseReadinessResult

MuseReadinessResult

Property	Type	Description
`status`	`MuseLaneStatus`	The `submission-ready` rollup — the headline verdict. `"ready"`, `"not-ready"`, or `"incomplete"`.
`technicalStatus`	`MuseLaneStatus`	The `technical-preflight` rollup, answerable without a submission profile.
`summary`	`string`	Human-readable rollup of `status`, naming what is missing when incomplete.
`context`	`MuseReadinessRunContext`	Target URL, declared capabilities, and evidence sources.
`lanes`	`MuseReadinessLaneResult[]`	Per-lane results (see Lanes).
`findings`	`MuseReadinessFinding[]`	Every graded statement about the target.
`classificationSheet`	`MuseClassificationRow[]`	Suggested Read / Write / Sensitive-write class for every listed tool. Empty when no listing was captured.
`policySnapshotDate`	`string`	ISO date of the policy corpus this run graded against.
`engineVersion`	`string`	Engine version stamp.
`startedAt`	`string`	ISO timestamp when gathering started.
`durationMs`	`number`	Total duration in ms.

Verdicts

`status`	Meaning
`"ready"`	Every requirement this run could evaluate is satisfied.
`"not-ready"`	At least one requirement is unmet.
`"incomplete"`	Nothing failed, but at least one requirement could not be evaluated — typically because no submission profile was supplied.

Lanes

Muse readiness runs four lanes. Each answers a different question.

Lane	What it checks	Can fail a verdict?
`runtime-compatibility`	HTTPS reachability and redirect behavior	Yes
`tool-policy`	Tool listing against §3–§4 (no credential in listing, combined read/write tools classified as write)	Yes
`submission-artifacts`	The §5 submission profile: overview, contacts, attestations, tool documentation, credentials, test account, and per-tool classifications	Yes
`experience-insights`	Heuristics over names and descriptions (sensitive-write signals, concealed writes, money-movement, trade tools)	Never — advisory only

Two staged verdicts

Stage	Lanes rolled up	When answerable
`technical-preflight`	`runtime-compatibility`, `tool-policy`	Always — no profile needed
`submission-ready`	`runtime-compatibility`, `tool-policy`, `submission-artifacts`	Requires a submission profile; `incomplete` without one

formatMuseClassificationSheet()

Renders the classification sheet as the Markdown table §5.6 asks submitters to include in their documentation.

formatMuseClassificationSheet(
  rows: readonly MuseClassificationRow[]
): string

const markdown = formatMuseClassificationSheet(result.classificationSheet);
// | Tool | Classification | Basis |
// | --- | --- | --- |
// | `search_products` | Read | declared; matches the suggestion |
// | `place_order` | Sensitive write | suggested: name contains "order"; ... |

The table uses the declared class from the submission profile where one exists, and the suggested class otherwise. A declared class that disagrees with the suggestion shows both, so a wrong row is easy to spot and overrule.

Submission profile

Supply a submission profile to grade the §5 requirements. All fields are optional — omitting a field leaves the corresponding checks not-evaluated rather than failing them.

const evidence = await gatherMuseReadinessEvidence({
  enteredUrl: "https://your-server.com/mcp",
  fetchFn: fetch,
  submissionProfile: {
    overview: "A product search and ordering connector for Acme Store.",
    contacts: [{ name: "Jane Smith", email: "jane@example.com" }],
    toolClassifications: {
      search_products: "read",
      place_order: "sensitive-write",
    },
    testAccount: {
      instructions: "Use the sandbox credentials in the shared vault.",
    },
    readOnlyOption: false,
  },
});

MuseSubmissionProfile fields

Field	Type	Description
`overview`	`string`	Short description of what the connector does.
`contacts`	`Array<{ name: string; email: string }>`	Submission contacts.
`attestations`	`string[]`	Attestation identifiers the submitter has agreed to.
`toolDocumentation`	`Record<string, string>`	Per-tool documentation keyed by tool name.
`integrationCredentials`	`object`	Credentials for Muse’s review team.
`readOnlyOption`	`boolean`	Whether the connector offers a read-only mode.
`testAccount`	`object`	Test account details for end-to-end QA.
`toolClassifications`	`Record<string, MuseToolClass>`	Declared Read / Write / Sensitive-write class per tool.

MuseToolClass is "read", "write", or "sensitive-write".

Classification sheet

The classification sheet is a suggested Read / Write / Sensitive-write class for every listed tool. It is a starting point for the §5.6 documentation table — Meta classifies each tool itself during review.

import { buildMuseClassificationSheet } from "@mcpjam/sdk";

const sheet = buildMuseClassificationSheet(tools, {
  place_order: "sensitive-write", // declared overrides
});

Each MuseClassificationRow has:

Property	Type	Description
`tool`	`string`	Tool name.
`suggested`	`MuseToolClass`	Suggested class based on annotations and heuristics.
`basis`	`MuseClassificationBasis`	Why the base class was chosen: `"annotation"`, `"combined-operations"`, or `"default"`.
`reasons`	`string[]`	One line per signal behind the suggestion.
`declared`	`MuseToolClass \| undefined`	The submitter’s declared class, when a profile supplied one.

Claude readiness CLI — readiness check claude and readiness start claude
OpenAI readiness CLI — readiness check openai and readiness start openai
Protocol Conformance SDK — MCP protocol checks
Host Compat SDK — host compatibility evaluation

Get Started

Concepts

Reference

Import

Quick example

gatherMuseReadinessEvidence()

GatherMuseReadinessEvidenceOptions

Supplying a pre-fetched tool listing

gradeMuseReadiness()

MuseReadinessResult

Verdicts

Lanes

Two staged verdicts

formatMuseClassificationSheet()

Submission profile

MuseSubmissionProfile fields

Classification sheet

​Import

​Quick example

​gatherMuseReadinessEvidence()

​GatherMuseReadinessEvidenceOptions

​Supplying a pre-fetched tool listing

​gradeMuseReadiness()

​MuseReadinessResult

​Verdicts

​Lanes

​Two staged verdicts

​formatMuseClassificationSheet()

​Submission profile

​MuseSubmissionProfile fields

​Classification sheet

​Related

Import

Quick example

gatherMuseReadinessEvidence()

GatherMuseReadinessEvidenceOptions

Supplying a pre-fetched tool listing

gradeMuseReadiness()

MuseReadinessResult

Verdicts

Lanes

Two staged verdicts

formatMuseClassificationSheet()

Submission profile

MuseSubmissionProfile fields

Classification sheet

Related