--- title: "doc0: Codebase Map for Teams and AI Agents" description: "doc0 maps your GitHub repo into a code knowledge graph and generates a wiki where every claim cites the code. Query both over MCP from Claude Code, Cursor, Slack, and Linear." last_updated: "2026-09-01" canonical_url: "https://www.doc0.app/" --- # doc0: Codebase Map for Teams and AI Agents > doc0 maps your GitHub repo into a code knowledge graph and generates a wiki where every claim cites the code. Query both over MCP from Claude Code, Cursor, Slack, and Linear. ## What doc0 does - Connect a GitHub repository. doc0 parses it into a symbol graph (functions, call chains, API endpoints) and generates structured documentation from that structure, with citations back to the source files. - The wiki is yours: edit pages, publish them on your own domain, and keep your edits when the docs regenerate as you push. - One repository gives you technical docs and a user guide, each published separately. - Every project ships an MCP server, so AI coding agents query the wiki and the code graph directly instead of guessing from raw files: search, page reads, symbol context, impact analysis, path tracing between symbols, open PRs likely to collide, and who knows an area of the code. - A Slack bot and a Linear agent answer from the same index (Team plan). - Generation and search embeddings run in the EU. Your code is analyzed in memory and never used to train models. ## For AI agents - MCP endpoint per project: `https://www.doc0.app/api/mcp//mcp` (Streamable HTTP). Call `doc0_get_minimal_context` first, then search or list pages, then read a page. - Public projects need no account or API key. Private projects use a project key (`doc0_proj_*`) or a personal key (`doc0_mcp_*`). - Every public wiki has a page index at `https://www.doc0.app/docs//llms.txt`, and every wiki page has a markdown twin: append `.md` to the page URL or send `Accept: text/markdown`. - Site index: https://www.doc0.app/llms.txt. Server card: https://www.doc0.app/.well-known/mcp.json. OAuth metadata: https://www.doc0.app/.well-known/oauth-protected-resource. How to authenticate: https://www.doc0.app/auth.md. - `npx doc0 setup` installs the MCP entry, an always-on rule and a project skill for Claude Code, Cursor or Codex; `npx doc0 remove` undoes it. ## Pricing Prices are in EUR. Every plan starts with a 3-day trial. - Pro: €24 per month billed annually, €30 billed monthly. For one person with private repos. - Team: €200 per month billed annually, €250 billed monthly. Unlimited members, plus the Slack bot and the Linear agent. ## FAQ ### How is this different from Mintlify or GitBook? Mintlify and GitBook are great editors, but you still write the docs yourself, and their AI features just summarize prose. doc0 parses your code into a symbol graph (functions, call chains, API endpoints), generates docs from that structure, and exposes both wiki and MCP from the same source. AI tools query the graph directly, not paraphrased prose. No manual writing, no MDX files, no CI pipeline to maintain. ### How is this different from DeepWiki? DeepWiki generates a one-time snapshot. doc0 keeps the docs current: your docs update when you push, your edits survive regeneration, and the same repo gives you both technical docs and user guides. Plus private repos, team collaboration, custom domains, and an MCP server your AI tools query directly. In our published benchmark, blind judges scored doc0 ahead of DeepWiki on every one of 10 repos tested. Methodology and raw scores are public. ### How does doc0 work with Claude Code, Cursor, Slack, and Linear? Your AI tools stop guessing about your codebase. Every doc0 project ships with an MCP (Model Context Protocol) server, scoped by a project key. Add it to Claude Code or Cursor with one line of config. MCP queries return grounded answers with source citations to the underlying code files. Beyond search, agents get structural tools raw file access can't offer: impact analysis, path tracing between symbols, detection of open PRs likely to collide, and who-knows expertise mapping. The Slack bot and Linear agent answer from the same index and are available on the Team plan: install the doc0 app in Linear and mention it on any issue, or ask the Slack bot from any channel. ### How do I make my agent actually use doc0? Installing an MCP server doesn't mean an agent uses it. Left alone, agents drift back to searching raw files. Run npx doc0 setup in your repo: it detects Claude Code, Cursor, or Codex and installs three things per agent: the MCP server entry, an always-on rule that points the agent at your docs first, and a skill with triggers tuned to your project. npx doc0 remove undoes exactly that. Public projects need no account or API key; private projects use your dashboard API key. ### Can non-engineers (product managers, support, customer success) use this? Yes. doc0 generates two documentation modes from the same code: a technical wiki for engineers and a user-guide wiki written for non-engineers. Both are searchable with AI chat on every plan. On the Team plan, that same context reaches non-engineers where they already work: product managers query the Linear agent when scoping tickets, and support and customer success ask the Slack bot. No MDX, no markdown skills required. The interface is built for reading, not authoring. ### Does it work with private repositories? Yes. doc0 connects via GitHub App. Your repository is cloned, analyzed in memory, and deleted after processing. We don't keep a copy. To power search and chat we store indexed chunks of your code (encrypted at rest, deleted with your project) plus file hashes for change detection, and your code is never used for training. Private repositories are available on Pro, Team, and Enterprise. ### Why not just run it locally? Local code-graph tools re-index on every machine and only help the one engineer running them. doc0 is hosted (no per-machine indexing), always-fresh (it auto-updates as you push), and team-shared (one graph powers Slack, Linear, and every AI tool your team uses). On top of the graph, you also get a human-readable wiki, something a local indexer doesn't produce. ### What happens to my edits when docs regenerate? Your manual edits are preserved. doc0 tracks provenance: which source files generated which pages. When you push code, only pages whose source files changed are regenerated. Edited pages are flagged but never overwritten. ### What languages and frameworks are supported? doc0 works with any GitHub repository regardless of language. It parses 14 grammars with tree-sitter (JavaScript, TypeScript, TSX, Python, Java, C, C++, C#, Go, Rust, Kotlin, PHP, Ruby, and Swift) and falls back to file-structure analysis for the rest. It detects endpoints in Express, NestJS, FastAPI, Django, Spring Boot, Rails, and 10+ more frameworks. ### How does the AI chat work? The chat retrieves the most relevant snippets from your docs and source code, then asks an LLM to compose a grounded answer with citations. Under the hood: hybrid vector + full-text search, query expansion to catch synonyms, and reranking to filter noise. You don't need to think about any of that. Every answer cites the files and symbols it's based on, so you can verify. ### How is the knowledge graph used? doc0 doesn't just summarize files. It maps your entire codebase into a symbol graph with call relationships, community detection, and cross-module execution flows. This powers 'How It Works' pages that trace real code paths, enriches AI chat with structural context, and surfaces authorship and change frequency for every file. ### How much does it cost? Pro and Team start with a 3-day trial. Pro (€30/mo) covers 500,000 lines of code a month, unlimited public projects, 5 private repos, custom domains, and export. Team (€250/mo) adds unlimited team members, 20 private projects, deep research, plus the Slack bot & Linear agent. Enterprise is priced per organization. Book a call. All plans run on EU-resident infrastructure. ### Why not just give Claude Code or Cursor my repo directly? Honestly: on a small, well-organized repo, a bare agent does fine. doc0 earns its keep when the repo outgrows a context window, and the moment anyone besides that one engineer, human or agent, needs the same answers. Agents reading raw files re-derive your codebase's structure from scratch every session, burning tokens and inventing function names when context runs out. doc0 builds the structure once (symbols, call graphs, execution flows) and serves it to every tool over MCP. The same index also answers your PM in Linear and your support team in Slack. Raw repo access can't do that. ### Does doc0 stop my agent from writing duplicate code? It makes duplicates findable before they're written. Agents query the knowledge graph over MCP and get the existing implementation with its callers, so "write a new helper" becomes "use the one that exists." doc0 doesn't gate your agent's writes (no tool truthfully claims to), but an agent that sees the codebase's structure stops recreating it. ### Where does my code get processed? Projects generate in the EU. Your repository is cloned, analyzed in memory, and deleted after processing. We keep file hashes, the indexed chunks that power search and chat, and their embeddings, never a copy of your repository. Your code is never used for training. Search relevance ranking uses a third-party processor; our full sub-processor list is in the privacy policy. ### How do you measure documentation quality? We publish our benchmark: blind LLM judges score doc0-generated wikis against DeepWiki and Google CodeWiki on 10 real repositories. Methodology, corpus, and raw scores are public in the benchmarks repository. ## Sitemap See the full [sitemap](https://www.doc0.app/llms.txt) for all pages, and the [public wiki gallery](https://www.doc0.app/explore). - [Privacy](https://www.doc0.app/privacy) - [Terms](https://www.doc0.app/terms)