Context Repo Documentation

Welcome to Context Repo

Context Repo is a context repository for AI agents and humans, a single home for the prompts, documents, and collections you reuse across every AI tool you already use. The documentation here covers everything from a five-minute install to the deep mechanics of the 29-tool Model Context Protocol (MCP) server, the REST API, the dashboard, the Chrome extension, and the ChatGPT App.

These docs are written for two audiences in equal measure. Humans setting up Context Repo for the first time get task-oriented guides with screenshots, copy-paste commands, and one-click install paths. AI agents and crawlers get a clean H1→H2→H3 structure, an /docs/llms.txt section pack, a markdown .md variant of every resource article, and explicit .well-known/ discovery surfaces so the same content is reachable through the channel that fits the consumer.

The fastest path to a working install is the Quickstart: sign up, save one prompt, connect your AI client, and run your first find_items call in under five minutes.

What's in these docs

How Context Repo works

Three primitives sit at the centre of the product: prompts (versioned templates with ${variable} placeholders), documents (text artifacts parsed from 75+ source formats and chunked at three hierarchy levels), and collections (named groups that scope retrieval and access). Every primitive belongs to one authenticated user and carries its own version history on every content edit.

Four surfaces reach into the same repository and read from the same source of truth:

• The dashboard at contextrepo.com/dashboard for humans editing content directly.
• The MCP server at contextrepo.com/mcp with 29 tools (27 power tools plus the OpenAI Apps SDK search and fetch aliases) for any Model Context Protocol client.
• The REST API under /v1 for programmatic integrations that prefer plain HTTP.
• The Chrome extension for one-click capture from any open browser tab.

For the longer-form architecture walkthrough, see the resources cluster: What Is an AI Context Repo for Agents? and Semantic Search and Deep Search: Two Retrieval Layers.

Standards and protocols

Context Repo is built on open standards so the same context works across every AI tool, today and as the ecosystem grows:

• Model Context Protocol over streamable HTTP transport with JSON-RPC framing. The MCP server advertises all 29 tools at /.well-known/mcp.json and a full tool catalogue at /.well-known/mcp/server-card.json.
• OAuth 2.1 with PKCE for the MCP transport. The authorization-server metadata follows RFC 8414 and the protected-resource metadata follows RFC 9728, published at /.well-known/oauth-authorization-server and /.well-known/oauth-protected-resource/mcp.
• OpenAPI 3.1 for the REST API. The machine-readable spec is at /openapi.json and a YAML mirror is at /api/openapi.yaml. See /docs/api/endpoints for the full reference and request/response examples.
• OpenAI text-embedding-3-small at 1536 dimensions for every prompt, document, and chunk in the index. The vector index is pinned to that dimension, so retrieval stays consistent across queries and ingest.
• Five authentication modes at the HTTP layer: Clerk session JWT, OAuth access token (oat_*), machine-to-machine token (mt_*), Clerk API key (ak_*), and first-party API key (gm_*). See /docs/api/authentication for setup and scope rules.

Documentation for AI agents

If you are an AI agent or a crawler reading these docs on behalf of someone else, here are the machine-readable surfaces designed for you:

• /docs/llms.txt: section-scoped llms.txt for the documentation subtree (this set of pages).
• /llms.txt: site-wide context pack covering the marketing site, dashboard, MCP server, REST API, and resources cluster.
• /llms-full.txt: the single-fetch full agent context pack.
• /.well-known/mcp.json: MCP server capability advertisement (SEP-1649).
• /.well-known/agent-card.json: A2A agent skills catalogue.
• /AGENTS.md: public agent-instructions file for any agent touching this codebase.
• /resources/<slug>.md: plain-markdown variant of every resource article, with Mermaid diagrams converted back to fenced ```mermaid blocks so agents see clean markdown instead of MDX JSX.

Use the surface that matches the consumer: the streamable HTTP MCP transport for live tool calls, the REST API for programmatic integration, the llms.txt packs for grounding, and the HTML docs for human readers.