Serving and connections

The endpoint

Address: https://baltor.ai/mcp
Protocol: Model Context Protocol (MCP)
Transport: Streamable HTTP
Sessions: Stateless: each request stands on its own
Each request: Carries its own key, as a bearer header

One authenticated endpoint serves every harness: Claude Code, Codex, OpenCode, Pi, the Baltor Harness and 17 more MCP tools.

Versions, on purpose

The service serves MCP versions 2025-11-25 and 2026-07-28. A request that names another version is answered with the versions the service serves, so your client chooses one. Your client picks the version, every time.

The key header

Each request carries your Baltor key in the Authorization header, read from BALTOR_SERVICE_TOKEN in your environment. The configuration holds the variable name; your key stays in your environment.

Bash

# List the tools your key can use
curl https://baltor.ai/mcp \
  -H "Authorization: Bearer $BALTOR_SERVICE_TOKEN" \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

OAuth connections

Harnesses that support OAuth connect at the same address. You approve each connection in your browser, on a Baltor page that names the app and its return destination, and its access stays within the scopes you allow.

Flow: Authorization code with PKCE (S256)
Refresh: Supported
Consent: Asked for every connection
Resource: https://baltor.ai/mcp

Scope	What it allows
`provisioning:metadata`	Metadata and search
`provisioning:read`	Bodies and downloads
`usage:read`	Your usage record
`billing:manage`	Billing sessions

Ask the service first

The capabilities record is the source of truth for everything on this page. Ask it before you plan a large job; it answers to anyone, key-free.

Bash

# Ask the service what it serves
curl https://baltor.ai/api/v1/capabilities

It names the protocol versions, the transport, the search modes, the download limits and the runtime, which read cloudflare-workers-d1-r2 on October 25, 2026.

Limits

Read from https://baltor.ai/api/v1/capabilities on October 25, 2026.

Limit	Value	Field in the record
References per search	50	`search_results`
Operations at once, per account	4	`concurrent_operations_for_each_account`
Operations at once, whole service	8	`concurrent_operations`
Request size	64 KiB	`request_bytes`
Response size	16 MiB	`response_bytes`
Body returned inline	16 KiB	`inline_body_bytes`
One download	1 MiB	`download_bytes`
Refused sign-in attempts, per address	30 in 60 seconds	`failures_allowed`, `window_seconds`
OAuth requests, per service process	240 a minute	`requests_per_minute_per_process`
OAuth token requests, per process	120 a minute	`token_requests_per_minute_per_process`
OAuth registrations, per process	10 a minute	`registrations_per_minute_per_process`

REST routes

Capabilities and health answer to anyone; every other route takes a Baltor key.

Route	What it does	Access
`GET /api/v1/capabilities`	What the service serves	Open
`GET /api/v1/health`	Health readings	Open
`GET /api/v1/session`	Your current session	Baltor key
`POST /api/v1/retrieval`	Search the library	Baltor key
`POST /api/v1/provisioning`	An item's manifest and files	Baltor key
`POST /api/v1/download`	Download a body	Baltor key
`GET /api/v1/usage`	Your usage record	Baltor key

Remove a connection

Revoke the key on Keys and connections. Revocation takes effect on the key's next request, and files already downloaded stay on your computer.

Bash

# After you close your harness
unset BALTOR_SERVICE_TOKEN

Need a hand? Status, the connection checklist and the operator's email address are on one page.

Get support