# Bring Trilocore into your AI workflow. > Connect a compatible AI client to Trilocore MCP over Streamable HTTP with your API key to inspect contract behavior, read traces, and run supported tests on isolated forks. 1. [Platform](https://trilocore.com/platform/) 2. MCP integration Connect a compatible AI client to Trilocore to inspect contract behavior and run supported tests on isolated forks using your authorized account. [Connect MCP](https://trilocore.com/integrations/mcp#quickstart) [Create API key in console](https://app.trilocore.ai/) Connection details - Status: Available - Endpoint: `https://mcp.trilocore.ai/mcp` - Transport: Streamable HTTP - Authentication: `Authorization: Bearer jns_...` - Required access: Trilocore account & API key permissions for requested workflow On this page - [Quickstart](https://trilocore.com/integrations/mcp#quickstart) - [Supported workflows](https://trilocore.com/integrations/mcp#capabilities) - [Example requests](https://trilocore.com/integrations/mcp#examples) - [Security & boundaries](https://trilocore.com/integrations/mcp#boundaries) - [Troubleshooting](https://trilocore.com/integrations/mcp#troubleshooting) Quickstart ## Connect a compatible MCP client in five steps Trilocore hosts a remote Model Context Protocol (MCP) endpoint over Streamable HTTP. You do not need to run a local node or install a local adapter when your AI client supports remote Streamable HTTP servers with custom request headers. 1. 1. Create a scoped API key sign in at [app.trilocore.ai](https://app.trilocore.ai/) and mint a dedicated `jns_...` key with permissions for your workflow 2. 2. Add the remote server URL point your Streamable HTTP MCP client at `https://mcp.trilocore.ai/mcp` 3. 3. Configure private Bearer auth set `Authorization: Bearer jns_...` in user-level secret settings — never commit keys to repository files 4. 4. Run a read-only check list tools and query supported capabilities to confirm authentication and scopes 5. 5. Run an isolated-fork workflow open an isolated fork, inspect returned state or trace evidence, and close the fork when done mcp.trilocore.ai/mcp · streamable httpIllustrative Streamable HTTP — `https://mcp.trilocore.ai/mcp` with `Bearer jns_...` View configuration details // Connection parameters for clients supporting Streamable HTTP + custom headers URL https://mcp.trilocore.ai/mcp Transport Streamable HTTP Header Authorization: Bearer jns_ // Example: user-scoped CLI registration (keeps token out of project .mcp.json) claude mcp add --transport http -s user trilocore \ https://mcp.trilocore.ai/mcp \ --header "Authorization: Bearer jns_..." Protocol endpoint — **not a browser web page**; unauthenticated GET requests return HTTP 401. Supported workflows ## Inspect contract state, traces, and isolated forks from your client Every tool call executes upstream under your authenticated Trilocore account and returns structured, size-bounded evidence to your connected client. ### Read contract state & live authority Inspect contract balances, raw storage slots, and resolved proxy, owner, or role relationships on an isolated fork pinned to a specific block. - Determinism: Pinned block height for reproducible reads - Output: Structured JSON with large values preserved accurately [Architecture Explorer →](https://trilocore.com/platform/architecture-explorer) ### Run supported tests off-mainnet Open an isolated EVM fork session, run read calls or test transactions, snapshot and revert state, compile or deploy test contracts where permitted, and replay bounded step sequences. - Safety: Isolated fork state only — never broadcast to live chains - Idempotency: Caller request IDs prevent duplicate mutations on retry [Fork environments →](https://trilocore.com/platform/fork-environment) ### Inspect traces & operation outcomes Retrieve call traces, revert outcomes, and paged recording steps for transactions executed in your fork workflow so your client reasons from actual execution rather than guesswork. - Outcomes: Reverted calls reported explicitly alongside trace data - Large payloads: Paged through short-lived private session artifacts [Auditing IDE →](https://trilocore.com/platform/auditing-ide) Example requests · Illustrative ## Prompts to try once your client is connected The prompts below are illustrative examples of questions you can ask a compatible AI client after connecting it to `https://mcp.trilocore.ai/mcp` and supplying the target chain, block, contract address, or transaction details. Example 01Trace analysis“Explain why this transaction reverted using the available trace.” Example 02Pinned state read“Read this contract’s state on an isolated fork.” Example 03Before / after diff“Compare the result before and after this test transaction.” Security, privacy, and scope ## Know what Trilocore MCP does and does not do Clear operational and privacy boundaries keep automated tool workflows predictable for security teams. ### Client compatibility & auth Requires an MCP client that supports remote Streamable HTTP and a custom `Authorization: Bearer jns_...` header. OAuth sign-in, dynamic client registration, one-click installers, and universal compatibility across every AI client are not offered. ### Focused fork & analysis scope Exposes isolated EVM fork sessions, state reads, traces, and architecture inspection. It does not expose every [Radar](https://trilocore.com/platform/radar), [Garudon](https://trilocore.com/platform/garudon), or [Protect](https://trilocore.com/platform/protect) capability, and it never signs or broadcasts live-network transactions. Use a dedicated API key so browser and MCP sessions do not contend on the same fork. ### External client data handling Tool outputs returned over MCP go to the external AI client you chose and are processed under that client’s own terms — separate from in-console [Janus AI](https://trilocore.com/platform/janus-ai) (see our [Privacy Policy](https://trilocore.com/privacy#ai)). This hosted service is also separate from `trilocore.com`’s unauthenticated read-only WebMCP documentation helper. Troubleshooting ## Common connection responses #### HTTP 401 Unauthorized The request omitted `Authorization: Bearer jns_...`, or the API key is malformed or revoked. Keys are shown once at creation in [app.trilocore.ai](https://app.trilocore.ai/); mint a replacement key if needed. Opening `https://mcp.trilocore.ai/mcp` directly in a browser tab also returns 401 by design. #### HTTP 403 Forbidden (insufficient scope or origin) The API key does not grant permissions for the requested read or fork-write operation, or the client sent a disallowed browser `Origin` header. Create a key with the required permissions for your workflow. #### Active workflow or session conflict One fork workflow runs at a time per session binding, and timed-out mutations pause new writes until inspected. Check current fork status or close the previous fork explicitly before opening a new one, and avoid modifying the same fork simultaneously from the web console. #### HTTP 429 Too Many Requests Concurrent MCP sessions per user are capped. Close idle client sessions or wait for the `Retry-After` interval before opening a new connection. ## Ready to connect your AI client? Create a scoped `jns_` API key in the console, or book a technical walkthrough with our engineering team. [Launch console](https://app.trilocore.ai/) [Explore Janus AI](https://trilocore.com/platform/janus-ai) [Book a demo](https://trilocore.com/demo) --- [Canonical HTML version](https://trilocore.com/integrations/mcp)