CodeWeave MCP
Provides tools to inspect, edit, and validate a Git repository, including context fetching, file reading, patch preparation and application, and running allow-listed tasks.
Allows connection to ChatGPT for AI-assisted coding.
This server does not currently have a deployable release on Glama. Follow the repository's setup instructions to install and configure it for your MCP client, including any required credentials and local dependencies.
CodeWeave
CodeWeave is a local-first MCP server that gives an AI coding client precise search, guarded edits, Git operations, semantic intelligence, and bounded Bash execution inside one repository.
It is built for developers who want capable coding tools without silently switching repositories, hiding partial edits, or sending an entire machine to a remote service.
Quick start
You need Rust, Git, and Bash. Windows users can use the Bash bundled with Git for Windows.
cargo install --git https://github.com/abhij1306/codeweave --locked
codeweave installThe interactive installer:
asks which repository CodeWeave should serve;
recommends local stdio or configures loopback HTTP;
creates
config.jsonand a private origin token;checks Git, Bash, the workspace, configuration, token permissions, port, and eager index startup;
prints a ready-to-paste MCP client command.
For automation, skip the wizard:
codeweave init --path /absolute/path/to/repository --config config.json
codeweave doctor --config config.jsonRelated MCP server: git-daily-review
Connect a client
Local clients: stdio
Stdio is the simplest and safest transport. The installer prints a block like:
{
"command": "/absolute/path/to/codeweave",
"args": ["serve", "--transport", "stdio", "--config", "/absolute/path/to/config.json"]
}Paste it into any MCP client that accepts a local command.
Loopback HTTP
codeweave serve --transport http --config config.jsonThe local endpoint is http://127.0.0.1:8813/mcp by default. HTTP requires the
bearer stored in .mcp-token; the token is a private origin credential, not an
LLM API key and not external user authentication.
Remote clients
Never publish the loopback origin by simply injecting its bearer into every public request. A public gateway must authenticate each external caller first, then add the private origin bearer for the hop to CodeWeave.
The bundled PowerShell helper uses mandatory HTTP Basic authentication at ngrok before it substitutes the origin bearer:
.\start-ngrok.ps1 -Config .\config.jsonPowerShell prompts for the public username and password. Use this helper only with clients that support Basic credentials. URL-only hosted connectors need an OAuth-capable MCP gateway instead. See ChatGPT setup and Claude setup.
What CodeWeave exposes
CodeWeave advertises one fixed 25-tool surface:
discovery:
workspace,code_retrieve,code_intelligence;guarded editing:
code_write,code_replace,code_replace_range,code_insert,code_delete,code_rename,code_preview,code_transaction;Git:
git_status,git_diff,git_log,git_show,git_blame,git_preflight,git_stage,git_commit,git_restore,git_push;commands:
bash,bash_status,bash_output,bash_cancel.
The repository is fixed for the process lifetime. All connected clients share the same workspace mutations, generation, and Bash runs.
Safety model
CodeWeave is privileged developer tooling. An authenticated client can run unsandboxed Bash as your OS account, so connect only clients and models you trust.
The server still enforces narrower invariants around that trust:
file reads and writes are capability-confined to the configured workspace;
snapshot hashes, range handles, and expected-content checks guard edits;
multi-file edits report partial application and recovery failures honestly;
Git path arguments and destructive operations use validation and explicit confirmation;
HTTP authentication happens before a bounded 4 MiB request-body read;
the origin accepts HTTP/1 only and caps concurrent connections;
bearer files use
0600on Unix and a protected explicit DACL on Windows.
Read SECURITY.md before exposing HTTP beyond loopback.
Configuration
codeweave install creates a strict configVersion: 2 configuration. Unknown
fields are rejected, including unknown nested fields. See
config.example.json for every option.
Useful commands:
codeweave doctor --config config.json
codeweave serve --transport stdio --config config.json
codeweave serve --transport http --config config.jsonOptional semantic backends are rust-analyzer, basedpyright-langserver, and
typescript-language-server. They are disabled until enabled in the generated
configuration.
Updating
Install the newest release from GitHub:
cargo install --git https://github.com/abhij1306/codeweave --locked --force
codeweave doctor --config /path/to/config.jsonOlder Windows releases created bearer files with inherited permissions. If
doctor rejects such a file, stop CodeWeave, remove only the configured token
file, and start it again so the token is rotated with a protected DACL.
Developing
git clone https://github.com/abhij1306/codeweave.git
cd codeweave
cargo fmt --check
cargo clippy --all-targets --all-features -- -D warnings
cargo test --all-targetsArchitecture and concurrency guarantees live in docs/architecture.md. Tool contracts are documented in docs/tools.md, and contributions are welcome through CONTRIBUTING.md.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
This server cannot be deployed
Maintenance
Related MCP Connectors
Git-backed platform for skills, tools, and context for AI agents
Code intelligence for LLMs. Analyze, search, and retrieve code from any public git repository.
AI-native git hosting — repos, PRs, issues, CI gates, and AI code review over MCP (60 tools).
Git collaboration for AI agents and people: tasks, hand-offs, reviews, teams, shared memory.
Related MCP Servers
- AlicenseBqualityDmaintenanceEnables AI assistants to perform code reviews by providing access to staged files, git diffs, and repository file content. It allows users to evaluate changes and context within any local git repository before committing or pushing.34 npmISC
- AlicenseNot gradedqualityBmaintenanceEnables AI agents to run automated daily code reviews, retrieve Markdown reports, and curate a project knowledge base across any Git repository.AGPL 3.0
- FlicenseNot gradedqualityCmaintenanceEnables autonomous coding agents to inspect Git commits, create pull requests, parse issue threads, and post reviews on GitHub repositories.-
- AlicenseNot gradedqualityBmaintenanceEnables AI agents to interact with a local-first git client that provides a knowledge graph of the repository, including graph analysis, issue and proposal management, and architectural review.2MIT