Transport
Slaunt is a standard Streamable HTTP MCP server, so MCP SDKs handle this section for you. Read it if you're writing a client by hand or debugging one. The parts that differ from a generic server are strict session binding and thread boundaries.
| Method | Path | Purpose |
|---|---|---|
POST | /mcp | JSON-RPC requests |
GET | /mcp with Accept: text/event-stream | Server notifications, mainly tools/list_changed |
DELETE | /mcp | End the session (204) |
Sessions
Send initialize without an Mcp-Session-Id header. The response header carries the new session ID, which you send on every request after that.
curl -i https://central-mcp-ccr-production.up.railway.app/mcp \
-H "Authorization: Bearer $TOKEN" \
-H "Content-Type: application/json" \
-H "Accept: application/json, text/event-stream" \
-d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-11-25","capabilities":{},"clientInfo":{"name":"my-agent","version":"1.0.0"}}}'Sessions are bound to the user and the client that created them, so a session ID can't be shared between tools or tokens. Reuse one session per connection: each new session can show up as a separate agent in Quaero.
| Case | Response | Fix |
|---|---|---|
| No header on a non-initialize request | 400, -32600 | Send the header |
| Session unknown, expired, or used with another token/client | 404, -32001, data.required_action: "initialize" | Re-initialize without the header |
| Session was deleted | 404, MCP session terminated (deleted)... | Re-initialize |
Notification stream
Open it if you cache tools/list: Slaunt sends notifications/tools/list_changed when the workspace's provider tools change.
| Status | Cause |
|---|---|
400 | No session header |
403 mcp_origin_not_allowed | Browser request from a disallowed origin |
409 | Too many streams open for this session |
429 | Server stream capacity exhausted; retry later |
Thread boundaries
By default one session equals one runtime. If your client runs several chats over one connection, tag each request with a boundary so Slaunt keeps them as separate threads instead of merging them. Put it in params._meta or a header. Values are strings up to 1,024 characters and are stored only as hashes. Conflicting values on one request are rejected.
| Boundary | _meta key | Header |
|---|---|---|
| Slaunt thread (preferred) | ai.slaunt/thread-instance | Ai-Slaunt-Thread-Instance |
| OpenAI session | openai/session | OpenAI-Session |
| Generic | mcp/thread-id | Mcp-Thread-Id, X-Conversation-Id |
| Session instance | ai.slaunt/session-instance | Ai-Slaunt-Session-Instance |
Without a boundary, each new chat must call thread_begin. Codex has its own routing.
Limits
| Limit | Value | When exceeded |
|---|---|---|
| Requests | 240/min | 429 http_rate_limited with Retry-After |
| Request body | 1 MiB | Rejected |
ccr_search query | 1,000 chars | Tool error |
Every response carries X-RateLimit-Limit, X-RateLimit-Remaining and X-RateLimit-Reset (Unix seconds).
Results
Each tools/call result has content[0].text, a readable summary meant for the model, and structuredContent, the full data meant for code. Failures come back as a JSON-RPC error, not as an isError result, so check for error first.