For buyers & agent developers

Connect one router instead of integrating many services.

The Tracert Router is a five-tool MCP server. Install it once and your agent can find an external service at the moment it needs one — compare real facts, take an exact quote, execute, and keep evidence it can re-verify later.

The five tools

Tool	What it does	Writes external state?
search_capabilities	Small candidate set for an intent + constraints, every match reason grounded in manifest facts.	No
get_capability	The full manifest: contract schemas, pricing, latency, data policy, evidence, provenance.	No
get_quote	Exact, expiring terms. Free capabilities quote an explicit zero — never a hidden handshake.	No
invoke_capability	Executes and returns a TRACE Receipt. Send an idempotency key; retries can never double-execute.	Yes
get_execution	Current receipt for any execution — poll async work, re-verify outcomes later.	No

Connect (hosted, read-only)

The hosted gateway is live at https://tracert.site/api/mcp — an MCP endpoint over Streamable HTTP. Point any MCP client at that URL:

{ "mcpServers": { "tracert": { "url": "https://tracert.site/api/mcp" } } }

The public gateway is read-only: free reads (exchange rates, dictionary, QR) execute live and return a TRACE Receipt with a reproducible output commitment. Capabilities that write external state are refused here — run the local router below to execute those.

Install (local — reads and writes)

The local router runs over stdio and can also execute writes (gated behind TRACERT_ENABLE_LIVE_SUBMIT=1). From the Tracert repository:

# build the router
cd router && npm install && npm run build

# register it with Claude Code (any MCP host works)
claude mcp add tracert -- node /path/to/tracert/router/dist/server.js

A plain HTTPS search API for non-MCP callers is next on the discovery roadmap.

Then try the benchmark prompt

"Submit this AI website to a directory that accepts programmatic
submissions. Do not pay anything, and return proof of the published page."

Your agent should search, pick the matching capability, read the contract, invoke, and hand you a receipt whose evidence (e.g. a pull request and a live page) you can check yourself. (Once a capability that satisfies this intent is published — the registry is a fresh slate today.)

Your policies stay yours

Budget: filter by free_only or max_price_usd; quotes are exact and expire.
Payment methods: filter to rails you can actually present (card-backed MPP, x402 wallet, prepaid, BYOK) — free routes need none.
Data: every manifest declares retention, training use, regions and subprocessors; exclude what your policy forbids.
Evidence freshness: require recent comparable executions before trusting a capability with real work.

What a receipt gives you

Every invocation returns a TRACE Receipt: the capability version and manifest hash, the quote you accepted, sha256 commitments over request and output, artifacts (like the live page URL), and evidence entries (repository records, HTTP observations). Completion is recorded separately from your acceptance — the network never pretends a technically-successful output is automatically a good one.

Browse capabilities /agents