glassBead-tc Audius MCP Server Test Suite
Run a comprehensive read-only test suite against the Audius MCP server by spawning parallel subagents. Each subagent tests one tool via real MCP tool calls (mcp__audius-mcp__*).
Prerequisites
The MCP server must be running and connected. Check that tools mcp__audius-mcp__search, mcp__audius-mcp__execute, mcp__audius-mcp__subgraph, and mcp__audius-mcp__play are available. If not:
- Build and start the server:
pnpm dev & - Add to
.mcp.json:{ "audius-mcp": { "type": "http", "url": "http://localhost:3000/mcp" } } - Reconnect MCP via
/mcp
Test Execution
Spawn 4 subagents in parallel (all in one message). Each subagent runs its test cases independently using the MCP tools directly — no curl, no JSON-RPC assembly. Each test should call the tool, check the result, and report PASS/FAIL.
Subagent 1: Search Tool Tests
Run these tests using the mcp__audius-mcp__search tool. Report each as PASS or FAIL with a brief result summary.
1. **No filters** — Call search with no arguments.
PASS if result mentions available tags and endpoint count.
2. **Tag: tracks** — Call search with tag="tracks".
PASS if results contain endpoints with "/tracks" in the path.
3. **Tag: users** — Call search with tag="users".
PASS if results contain endpoints with "/users" in the path.
4. **Query: trending** — Call search with query="trending".
PASS if at least one endpoint relates to trending.
5. **Method: GET** — Call search with method="GET".
PASS if all returned endpoints have method "GET".
6. **Path: playlists** — Call search with path="/playlists".
PASS if results contain "/playlists" in paths.
7. **Combined filters** — Call search with tag="tracks" AND method="GET".
PASS if all results are GET endpoints under tracks.
Save a summary to /tmp/audius-test-results/search-results.md
Subagent 2: Execute Tool Tests
Run these tests using the mcp__audius-mcp__execute tool. Report each as PASS or FAIL with a brief result summary.
1. **Trending tracks** — Execute: return await audius.request('GET', '/tracks/trending', { query: { limit: 3 } })
PASS if response contains track data with title and user fields.
2. **Search user** — Execute: return await audius.request('GET', '/users/search', { query: { query: 'deadmau5' } })
PASS if response contains user data.
3. **Chain calls** — Execute:
const trending = await audius.request('GET', '/tracks/trending', { query: { limit: 1 } });
const trackId = trending.data[0].id;
return await audius.request('GET', '/tracks/' + trackId)
PASS if response contains a single track's details.
4. **Console capture** — Execute: console.log('hello from sandbox'); return 42
PASS if output contains "hello from sandbox" AND return value is 42.
5. **Error handling** — Execute: return await audius.request('GET', '/nonexistent/endpoint/xyz')
PASS if response indicates an error.
6. **Multi-step** — Execute:
const users = await audius.request('GET', '/users/search', { query: { query: 'RAC' } });
if (!users.data || users.data.length === 0) return 'no users found';
const userId = users.data[0].id;
const tracks = await audius.request('GET', '/users/' + userId + '/tracks');
return { user: users.data[0].name, trackCount: tracks.data ? tracks.data.length : 0 }
PASS if response contains a user name and track count.
7. **Custom timeout** — Execute with code "return 'fast'" and timeout 5000.
PASS if response returns "fast" without timing out.
Save a summary to /tmp/audius-test-results/execute-results.md
Subagent 3: Subgraph Tool Tests
Run these tests using the mcp__audius-mcp__subgraph tool. Report each as PASS or FAIL with a brief result summary.
1. **Network stats** — Query: { audiusNetworks(first: 1) { totalTokensStaked totalTokensDelegated totalSupply } }
PASS if response contains numeric values for staking/supply.
2. **Service types** — Query: { serviceTypes(first: 5) { id isValid minStake maxStake } }
PASS if response contains service type entries.
3. **Recent proposals** — Query: { proposals(first: 3, orderBy: submissionBlockNumber, orderDirection: desc) { name outcome numVotes } }
PASS if response contains proposal data.
4. **Service nodes** — Query: { serviceNodes(first: 5, where: { isRegistered: true }) { endpoint type { id } } }
PASS if response contains node endpoints.
5. **Invalid query** — Query: { nonExistentEntity { id } }
PASS if response contains a GraphQL error message.
Save a summary to /tmp/audius-test-results/subgraph-results.md
Subagent 4: Play Tool + Cross-Tool Tests
Run these tests using the mcp__audius-mcp__play and mcp__audius-mcp__search tools. Report each as PASS or FAIL.
Note: The play tool opens a browser/app, which won't work in headless/remote environments. Test it anyway — a "failed to open" error with valid track data still counts as PASS for the API resolution part.
1. **Play by query** — Call play with query="Tiny Little Adiantum".
PASS if response contains track info (title, artist, url) regardless of whether the browser opened.
2. **Play by invalid ID** — Call play with trackId="INVALID_NONEXISTENT_ID_12345".
PASS if response indicates an error (track not found or API error).
3. **Play with no args** — Call play with no arguments.
PASS if response indicates an error about missing trackId or query.
4. **Search → Play integration** — First call search with query="trending", then call play with a query based on what you found.
PASS if play resolves the track successfully.
5. **Search endpoint count** — Call search with no args, verify the reported endpoint count is > 0.
PASS if count is a positive number.
Save a summary to /tmp/audius-test-results/play-cross-results.md
After All Subagents Complete
Collect all result files and produce a summary table:
## Test Results
| Group | Tests | Passed | Failed | Flaky |
|--------------|-------|--------|--------|-------|
| Search | 7 | ? | ? | ? |
| Execute | 7 | ? | ? | ? |
| Subgraph | 5 | ? | ? | ? |
| Play + Cross | 5 | ? | ? | ? |
| **Total** | **24**| **?** | **?** | **?** |
For failures, include the test name and what went wrong. For flaky results (rate limits, timeouts, network issues), note them separately.
Critical: Subagent Constraints
Every subagent prompt MUST include these instructions at the top:
IMPORTANT CONSTRAINTS:
- You are a TEST RUNNER. Your ONLY job is to call MCP tools and report results.
- Do NOT read, edit, or modify any source code files.
- Do NOT run pnpm build, pnpm dev, or any build/restart commands.
- Do NOT attempt to fix, debug, or diagnose issues in the codebase.
- If a test fails, report it as FAIL with the error message. That's it.
- The only tools you should use are mcp__audius-mcp__* tools and Write (for the results file).
This prevents subagents from going off-script and modifying the codebase when they encounter test failures.
Important Notes
- All tests are read-only. No uploads, no favorites, no mutations.
- Tests use real MCP tool calls through the connected
audius-mcpserver — not curl or raw HTTP. - The play tool's browser/app opening will fail in headless environments (Codespaces, CI) — that's expected. Test the API resolution part, not the OS integration.
- Tests hit the live Audius API and The Graph — they require internet access.
- If a test fails due to rate limiting or network issues, mark it FLAKY rather than FAIL.