every hosted tool — and the REST endpoint each one mirrors.
2 pages
MCP tools
All 35 tools below are callable by any MCP client connected to https://innernet.live/api/mcp (JSON-RPC 2.0, Authorization: Bearer <token>). Most users never call them directly — the AI client wraps them.
They live in one place: lib/mcp/registry.ts holds every tool's description, schema and handler, and the parity test fails if one of them lacks a REST twin. What you read here is what the server serves.
Every one of them has a REST twin. The two doors run the same handlers against the same memory — see the parity table on the API page, or GET /api/v1 for the machine-readable version. Use MCP if you are an AI client with a connector; use the REST API if you are an agent or a script.
The one door. Pass the user's ask as intent and innernet compiles a budgeted slice instead of the whole cabinet: SPINE (identity, core, what moved lately, as_of), FOCUS (the sections that match, in full), PERIPHERY (a one-line index of everything else, with how to expand it), and retrieval_advice — what to read next if the conversation turns.
innernet_session
What innernet has made of this conversation so far — the chapter, the map it is working against, what it kept, what is set aside.
List your context maps — slug, name, type, tagline, last update. Plus memories other people have shared into your rooms.
innernet_load_project
Open a map. Lean by default (a line per dimension); full:true for every body. Carries the capture protocol and a disclosure-gated slice of your Self Map.
innernet_save_context
Create/update a project. Upserts dimensions and nodes by name — merge, never wipe.
innernet_get_dimension
Read one dimension's markdown without loading the whole map.
Capture a note/decision/insight. Without project_slug the Concierge auto-routes it (project, Personal Memory, both, or discard) — and the raw content is persisted before any routing, so a routing failure can never lose it.
search
Ranked full-text search across dimensions, nodes, captures, projects, and memories. (Also drives ChatGPT Deep Research.)
fetch
A project's full content as one markdown blob (ChatGPT Deep Research).
innernet speaks MCP 2026-07-28 — the stateless revision — and every revision before it, so nothing you have connected today needs to change.
No handshake required. Call server/discover (it needs no token — you should be able to learn how to talk to us before you sign in), or skip it and put io.modelcontextprotocol/protocolVersion in each request's _meta. initialize still works for clients that haven't migrated.
Interactive views rendered inline in the conversation — one live innernet window rather than a stack of panels.
io.modelcontextprotocol/tasks
Long work returns a durable handle instead of a wait. Declare it in your _meta capabilities and innernet_self_sync answers with a taskId you poll via tasks/get.
A fold can ask you something. If your client also declares elicitation, a sync task may pause at input_required carrying one question the Concierge couldn't resolve on its own — who someone is to you, whether something is a standing preference. Answer it with tasks/update and it is never asked again. One question, at the moment innernet is already thinking about it, and only where your client can show it.