
aoci-code MCP Server: Setup for Claude Code, Cursor and Codex
Gives Claude Code, Codex, Cursor and OpenCode a saved, Git-tracked map of your codebase and database schema to read before they change anything.
aoci-code builds and keeps up to date a written map of your repository and its database schema, so your coding agent reads the map instead of re-reading the code on every task. You install a small program on your own machine and your agent (Claude Code, Codex, Cursor or OpenCode) talks to it over a local connection. Your agent writes the map entries itself, and the program checks them and saves them in your repo as Git-tracked files.
Setup is a binary plus an init step, not a package install, and the project is a release candidate under the FSL-1.1-MIT license, which is source-available rather than open source. Expect the first index to take a while on a large repo.
Good for
- Developers whose coding agent keeps re-reading the same large repo
- Teams who want the agent's understanding of the system saved in Git
- People with a PostgreSQL or MySQL schema the agent should know about
- Anyone happy to let the agent write and maintain a lot of index files
Works in
12 should work
Tap an app to see how to set it up and any known problems.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Client does not support this transport.
Client does not support this transport.
Install aoci-code MCP
Claude Code
- 1
Copy this prompt.
Add the aoci-code MCP server to Claude Code. Run this command: aoci --repo /absolute/path/to/repository init --agent claude Then check that it loaded and tell me if I need to sign in or restart.
- 2
Paste it into Claude Code. It shows the change and asks before running it.
Requires aoci binary at a stable absolute path, Go toolchain only if you build from source.
Read-only mode: There is no read-only mode.
Get the binary from the GitHub Releases page and follow the verification steps in the install guide before you run it. Run the command from your project, using the full path to the binary, then ask your agent to scan and build the index. Version v0.1.0-rc23 is a release candidate. This merges an aoci server into the project .mcp.json. Add --hooks only if you want the optional pre-write guard. Restart Claude Code afterwards.
From the project README. Source.
Tools
Loads the working rules the agent follows for this project
Hands the agent the project rules for how to read and update the index. The agent calls it when it has no reliable copy of the rules, and it is not meant to repeat within one session.
Try: “Confirm the AOCI server is connected and load its rules.”
Write ·
aoci_rulesGives the agent the whole map of your codebase in one read
Sends the full index of your project to the agent, in chunks when it is large, then asks the agent to confirm it received all of it. This is the call that gives the agent a picture of the system before it starts a task.
Try: “Load the AOCI overview for this repo, then explain how billing is structured.”
Write ·
aoci_overviewPulls up specific index entries for named files or folders
Fetches the index entries for particular files, folders or database tables. It is meant for filling a gap after the overview, not for replacing it.
Try: “Show me the AOCI entries for the internal/auth folder.”
Write ·
aoci_get_entriesSearches the index by keyword or tag
Looks through the index entries for a keyword or tag and returns the matches. The agent uses it when it does not yet know which file it needs.
Try: “Search the AOCI index for entries about retry logic.”
Write ·
aoci_searchReads the tag dictionary and entry-writing rules
Returns the vocabulary the index uses, such as the tag dictionary and example entries. The agent checks it when it is unsure how an entry should be written.
Try: “Show me the tag dictionary AOCI uses for this project.”
Write ·
aoci_headerFinds index entries that are out of date after code changes
Compares the index with the current code and returns the next batch of entries that are missing, stale or orphaned, for the agent to write. It also records receipts and a local audit entry, and it can advance the baseline for pure formatting changes.
Try: “I have finished this change and the tests pass. Bring the AOCI index back in line.”
Write ·
aoci_maintainWrites the agent's new or revised entries into the index
Saves a full batch of entries the agent wrote for the changed files. It checks them against the source and the tag rules first, then writes the index files and baseline in one step, or rejects the batch.
Try: “Write the AOCI entries for the files I just changed.”
Write ·
aoci_update_entryDeletes an index entry whose file or table no longer exists
Removes one index entry when the file or database table it described is gone. It does not delete any source file, database object or saved evidence, and it refuses entries that are not orphaned.
Try: “Remove the AOCI entry for the file I deleted, if it is flagged as an orphan.”
Delete ·
aoci_remove_entryRecords a follow-up note when the agent lacks enough evidence
Saves a note about work that needs a person or more evidence, plus an audit record, without touching the index. The agent uses it instead of guessing an entry it cannot support.
Try: “Log a follow-up that the payments module needs a human to confirm its purpose.”
Write ·
aoci_report
Read from the source code
Permissions
medium risk · 6 things to watch
Medium risk
It writes index files into your repository and can delete index entries, and its overview tool tells the agent to keep going without asking you. Review the changes it makes to your repo in Git, and check the binary against the signed release before you run it.
What it can do
- Read your data
- Create and change things
- Delete things
- Read-only mode
Watch out for
It writes index files into your repository
init adds aoci.txt, aoci.meta.txt, aoci.code.txt, AGENTS.md and a .aoci folder, and the agent updates them as it works. Review them in Git like any other change. github.com
It can delete index entries
The remove tool deletes an index entry only when its file or table is gone, and it removes no source file or database object. Check what the agent removes before you commit. github.com
Tool text tells the agent not to ask you
The overview tool tells the agent to fetch every chunk automatically and not to ask the user or give a partial answer. Keep that in mind if you want to approve each step. github.com
Release candidate, source-available license
The current release is v0.1.0-rc23 under FSL-1.1-MIT, which bars using it in a competing product or service. Follow the project's signature checks before you run the downloaded binary. github.com
Database indexing reads schema only
It runs only when you declare a database and run an explicit command, reads table structure and not rows, and runs no DDL or DML. The README names PostgreSQL, MySQL and openGauss. github.com
Optional AI endpoint makes network calls
AI is off by default. If you set up your own endpoint, ai test and doctor --net send real network requests to it. github.com
Security advisories
No published security advisories as of Nov 1, 2026.
Technical details
| Turning on read-only | No read-only switch is documented. The read tools can still append a local audit record. |
|---|---|
| Sign-in and keys | No account or API key. A database index needs a connection string in an environment variable you name; AOCI stores no credentials. |
Questions
Is the aoci-code MCP server official?
No. It is a community project from the aoci-spec GitHub organization, and it is not made by Anthropic, OpenAI or Cursor. It is not tied to one app, so there is no vendor server to compare it with.
Is aoci-code free and open source?
It is free to use but not open source. The license is FSL-1.1-MIT, which allows internal use, education and research, and forbids offering a competing product or service. The license names MIT as the future license.
Which apps does it work with?
Codex, Claude Code and OpenCode get their project config written by the init step. Cursor only gets a snippet printed that you paste into its config yourself, and Qoder is also covered. Other apps that run local MCP servers need manual setup.
Do I need an API key?
No. Your coding agent writes the index, so the server needs no key of its own. You can optionally point it at your own OpenAI-compatible endpoint, with the key held in an environment variable you name.
Does it work read-only?
No read-only switch is documented. Even the check commands can append to a local audit log, and the README says to use a copy of the repo if you need zero file writes.
Does it read my database?
Only if you set one up and run the database command yourself. It reads table structure, not rows, for PostgreSQL, MySQL (5.7.6 and later) and a limited openGauss profile.
How do I add it to Claude Code?
Download and verify the binary, then follow the Claude Code tab in the install box above. Restart Claude Code afterwards so it loads the new server.
Why do I have to restart the agent after init?
The agent session that ran init has not loaded the server init just wrote. Some apps load servers on the fly and need no restart.
Information
updated Nov 2, 2026
- Publisher
- Community
- Transport
- stdio
- Auth
- None
- Version
- 0.1.0-rc23
- License
- FSL-1.1-MIT
- Last commit
- Oct 31, 2026
- Open issues
- 21
Sources
Docs, issue trackers and discussions this page is built from. Issues, quotes and findings above link to their own source.
Show all 9Hide
- AOCI-CODE README (v0.1.0-rc23) · github.com
- AOCI-CODE agent integrations · github.com
- AOCI-CODE MCP tool list (golden file) · github.com
- aoci_overview description · github.com
- aoci_remove_entry description · github.com
- aoci_update_entry description · github.com
- aoci_report description · github.com
- aoci_maintain description · github.com
- FSL-1.1-MIT license · github.com