agent-config MCP server
Commit Oct 21, 2026Provides portable agentic engineering rules and an MCP tool server for managing project logs and global agent instructions.
Also known as anansio/agent-config
- Stars
- 0
- Tools
- —
- Downloads, 30 days
- —
- Latest version
- —
- Last commit
- Oct 21, 2026
- License
- MIT
What is agent-config MCP?
agent-config provides portable agentic engineering rules and an MCP tool server for managing project logs and global agent instructions.
Summary generated from the project's README
About agent-config MCP server
Maintaining consistent engineering standards and project history across different AI environments is what this tool enables. It functions as a portable Agent Plugin that provides a standardized set of global rules and helper tools for agentic workflows. Developers can load it into any client supporting the Agent Plugins 1.0 format to ensure their agentic engineering setup remains uniform regardless of the specific project or local environment.
The package includes a stdio MCP server that manages project logs using a JSON Lines storage system. It provides specific tools like log_append, log_update, log_delete, and log_search to track decisions, features, and fixes. By default, logs are stored in a .agents/log.jsonl file within the project directory, though users can specify custom paths. The server relies solely on the Python standard library, making it compatible wherever python3 is available.
Beyond logging, the server declares a remote Context7 documentation server and includes on-demand skills for code review, documentation rules, and git workflows. It also bundles baseline references for security, accessibility, planning, and coding to guide agent behavior.
What you can do with it
- 1Standardize agent behaviorEnsure your AI agent follows the same coding and security rules across every project you work on.
- 2Correct agent mistakesUpdate or remove previous log entries if the agent changes course or discovers a previous path was a dead end.
Written from the project's README. Every use case is checked against a quote from it.
README
View on GitHubagent-config
Global agent rules and a small MCP tool server, packaged as a portable Agent Plugin. Load it into any client that supports the Agent Plugins 1.0 format and your agentic engineering setup follows the same baseline everywhere.
The plugin keeps global rules and helper tools in one versioned place, so individual projects stay clean.
What is in here
agent-config/
├── plugin.json # portable manifest
├── mcp.json # stdio MCP server + remote Context7 declaration
├── skills/
│ ├── global-rules/ # always-on baseline rules
│ │ ├── SKILL.md
│ │ └── references/
│ │ ├── accessibility.md
│ │ ├── cleanup.md
│ │ ├── coding-rules.md
│ │ ├── logging.md
│ │ ├── planning.md
│ │ └── security.md
│ ├── code-review/ # on-demand skill: subagent review gate
│ ├── docs-rules/ # on-demand skill for documentation
│ └── git-workflow/ # on-demand skill
├── tools/
│ ├── mcp_server.py # stdio entry point (thin launcher)
│ └── agent_mcp/ # server package
│ ├── server.py # JSON-RPC loop and tool registry
│ ├── log_store.py # JSON Lines storage and entry ids
│ └── tools/ # one module per tool group
│ └── log.py
├── tests/ # server and plugin validation tests
├── .github/workflows/ci.yml # runs the tests on push and pull request
├── README.md
├── CHANGELOG.md
├── SECURITY.md
└── LICENSE
Skills
global-rules is meant to be loaded always: its description says so, so a
client that only loads a skill by its description still picks it up. It points
at the bundled references for planning, logging, security, cleanup,
accessibility, and coding. code-review, docs-rules, and git-workflow load
on demand from their descriptions.
The MCP server
The server lives in the tools/agent_mcp/ package: server.py holds the
JSON-RPC loop and the tool registry, log_store.py owns storage, and each tool
group is its own module under tools/agent_mcp/tools/. tools/mcp_server.py
is a thin stdio entry point. It uses only the Python standard library, so it
starts anywhere python3 exists.
Entries are stored as JSON Lines (one entry per line) with an id and a UTC timestamp, so appending stays O(1) and searching streams the file instead of loading it.
By default the log is written to .agents/log.jsonl in the project directory.
Every tool also takes an optional path, so a project can keep its log wherever
it likes. The file is plain text and can be committed with the project if you
want the work history to travel with it.
log_append- add a short entry: what was done and why, with an optionalkind(feature,fix,decision).log_update- rewrite an entry by id after a course change, and recategorize it with an optionalkind.log_delete- remove an entry that turned out to be a dead end.log_search- find entries by substring orkind; returns them with their ids.
Add more tools by extending the specs in the relevant group module under
tools/agent_mcp/tools/, then importing that group in server.py.
MCP servers
mcp.json declares two servers:
agent-tools- this repository's stdio server, bundled with the plugin.context7- the remote Context7 docs server athttps://mcp.context7.com/mcp. It needs no credentials; an optional API key only raises rate limits and is a client-side concern, so it is not declared here.
Anything requiring a secret (an API key or token) belongs in the client, not in
the plugin: the Agent Plugins format has no portable credential field, and
OpenHands' loader reads mcp.json literally, without ${VAR} expansion.
Tests
The suite uses only the standard library:
python -m unittest discover -s tests -v
tests/test_mcp_server.pycovers the log tools, including a full stdio round trip with a real tool call.tests/test_plugin.pyvalidates the plugin itself: the manifest, the MCP wiring (it starts the declared server), and each skill's frontmatter and relative links.
CI runs the suite on every push to main and every pull request, across Python
3.10 to 3.13.
Install
Register this repository as a plugin source in a client that supports the Agent Plugins format. With the default branch:
github:anansio/agent-config
See the Agent Plugins documentation for how to add a source in your client.
License
MIT. See LICENSE.
agent-config MCP questions
What tools are included in the MCP server?
The server includes tools for appending, updating, deleting, and searching log entries.
Where are the logs stored by default?
Logs are written to a .agents/log.jsonl file located in the project directory.
Does the Context7 server require credentials?
No, the Context7 remote docs server needs no credentials, though an optional API key can be used to raise rate limits.
How do I install this plugin?
Register the repository as a plugin source in a client that supports the Agent Plugins format using 'github:anansio/agent-config'.