Codna logo - MCP Servers AI tool

Codna

MCP ServerFree

Codna connects an AI assistant to Codna’s repository tools through the Model Context Protocol. It can map code to triage issues, plan bug fixes, classify security findings, recall code from local memory, and file feedback.

Last updated:

  • Official. Built and maintained by the company or project behind the service, so usually kept in step with it. Not audited by us.

  • Community. Built and maintained by an independent developer or group, not by the company behind the service. The company has not reviewed it, and updates depend on its author.

About Codna

Codna connects an AI assistant to Codna’s repository tools through the Model Context Protocol. Running as a local server, it exposes tools for understanding a codebase, investigating a reported bug, reviewing security scanner findings, recalling relevant code, and sending feedback to the project. This lets someone work with Codna from a supported MCP client rather than switching exclusively to the command line. The tools have different requirements and effects: several are read-only, while opening a pull request or filing a report can write to GitHub.

What can Codna do?

Codna exposes five tools. Repository triage maps a repository and identifies files relevant to an issue; it is deterministic, read-only, and uses zero model tokens. The fix tool analyzes a bug and returns a fix plan by default. It can open a pull request when supplied with a Git URL and a GitHub write token.

  • Classify SARIF scanner findings by reachability and report autofix eligibility.
  • Search code in local on-device memory without calling a model or changing the repository.
  • File a bug, feature request, or question to the project’s public feedback GitHub repository.

The security tool requires a SARIF file with complete provenance. It classifies and reports findings; remediation is performed through Codna CLI commands, not through that MCP tool. The feedback tool can return a pre-filled issue link when it lacks a suitable token for automatic submission.

How Codna connects and authenticates

The documented MCP server runs locally over stdio, with the MCP client starting it as a child process. The guide names Cursor and Claude Desktop and also documents use with another MCP client. There is no hosted MCP endpoint documented in the reviewed guide. Executing any tool requires one-time free Codna device authorization; listing tools and initialization are credential-free. The fix tool additionally needs a model-provider key, and automatic GitHub submissions require a GitHub token. MCP support is an optional install extra, not part of the base Codna package.

Who is Codna best for?

Codna suits developers who want repository triage, fix planning, security finding analysis, code recall, or project feedback available inside an MCP client. Its verified individual plan is Free and includes the MCP server. Users need free Codna device authorization to execute tools. Fix analysis also needs a provider key, with charges billed by that provider; Codna’s pricing page lists the MCP server under its free individual plan.

What Codna can access and change

Repository mapping and recall run locally; for model-assisted fixes, Codna sends an issue-specific evidence bundle to the selected provider, and secret redaction is always on. Triage, security classification, and recall are read-only. The fix tool does not apply its plan to a local checkout, but can open a pull request when given a Git URL and write-capable GitHub token. The feedback tool can create a real issue in the project’s public GitHub repository when authorized. Without a suitable token, it returns a pre-filled issue link instead.

Alternatives to Codna

For adjacent repository and developer workflows, GitHub Official MCP Server connects an assistant to GitHub, while Supabase MCP Server connects one to Supabase projects. These are alternatives to consider when the task centers on interacting with those services rather than Codna’s documented repository triage, fix planning, security classification, local recall, and feedback tools.

Key Features

Five tools for repository triage, bug-fix planning, security finding analysis, code recall, and feedback reporting.

Maps repositories and identifies issue-relevant files without model tokens.

Analyzes bugs and returns a fix plan; opening a pull request requires a Git URL and GitHub write token.

Classifies SARIF findings by reachability and reports autofix eligibility; does not remediate through the MCP tool.

Searches code in local on-device memory without calling a model or changing the repository.

Runs locally over stdio with Cursor, Claude Desktop, or another MCP client.

Tool calls require free Codna device authorization; codna_fix also needs a provider key.

Secret redaction is always on; model-assisted fixes send an issue-specific evidence bundle to the selected provider.

Before you installSecurity checklist for MCP servers

An MCP server runs code and gives your AI assistant access to files, accounts or data. A few minutes of checking is worth it.

  1. Check the source. Make sure the repository belongs to the vendor or author you expect, that the vendor's own documentation links to it, and that it is actively maintained. Beware of lookalike names.
  2. Read what it can do. Look at the tools the server exposes and the permissions or scopes it asks for. Start with read-only access where the server allows it.
  3. Use the least access that works. Give it a dedicated API key or a narrowly scoped token, ideally on a test project or sandbox, not your production credentials.
  4. Pin versions and review updates. Install a specific version, read changes before upgrading, and do not run copied commands you do not understand.
  5. Treat what it returns as untrusted. Documents, web pages and tool results can contain text that tries to steer your AI assistant. Keep confirmation switched on for actions that write, delete, send or spend.
  6. Know where your data goes. Find out whether the server runs on your machine or is hosted, and what it sends to the vendor or to third parties.
  7. Keep a way back. Know how to revoke the key and remove the server from your client's configuration.

This page is general information, not security or legal advice, and listing a server is not an endorsement or a security review. We do not create, host or control MCP servers, and listing details such as commands, permissions or pricing can be out of date, so check the official source. You install and use MCP servers at your own risk. See the section on third-party products in our Terms of Service.

Pricing Plans

Free

$0

  • Includes the MCP server for individuals using Cursor or Claude.
  • Repository triage, security analysis, and code recall run locally without model tokens.
  • Free Codna device authorization is required to execute tools.
  • Bug-fix analysis also requires a model-provider key, billed by that provider.

Pricing information last updated: October 20, 2026

FAQs

Does the MCP server run locally or through a hosted endpoint?

The documented server runs locally over stdio. The reviewed MCP guide does not document a hosted MCP endpoint.

Which clients are documented?

The guide names Cursor and Claude Desktop, and says the server can be used with another MCP client.

What authorization does it need?

Executing any of the five tools requires one-time free Codna device authorization. codna_fix also requires a model-provider key. Automatic report filing requires a GitHub token.

Can codna_fix change my local checkout?

No. It returns a fix plan by default and does not apply it to a local checkout. Opening a pull request requires a Git URL and a GitHub write token.

Can codna_secure fix scanner findings?

The MCP tool classifies SARIF findings and reports autofix eligibility. Remediation is performed through Codna CLI commands, not this MCP tool.