Okto Pulse MCP
Okto Pulse tracks ideas, specs, tasks and bugs on a gated board that agents operate through 285 MCP tools
- 137 ★ on the repository
- Last commit
Extended access
Why this level
- The agent gets 285 tools that create and change tasks, specs and bugs on the board and write to the local database and knowledge graph
- The agent key sits in `.mcp.json` inside the project and is easy to commit by mistake
Install
- Works in
Manual install
pip install okto-pulsePython 3.11 or newer required; then `okto-pulse init` and `okto-pulse serve`.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The agent gets 285 tools that create and change tasks, specs and bugs on the board and write to the local database and knowledge graph
- The agent key sits in `.mcp.json` inside the project and is easy to commit by mistake
This is third-party code. Review the repository files before installing.
What it does
A local workbench for development with AI agents. An idea moves through stories, ideation, refinement, spec, tasks, tests and bugs. Every stage has structured artifacts, lineage and validation rules. An agent creates and updates them through MCP, while a human inspects and steers the same work in the web UI. A card changes status only after 17 gate checks. A spec must cover acceptance criteria, requirements, business rules and API contracts, a task starts only with scenario coverage and closing needs a conclusion with completeness and drift assessment. An embedded per-board knowledge graph keeps decisions and lessons from past bugs, and search uses the `all-MiniLM-L6-v2` model.
Who it is for. Teams and solo developers on Claude Code, Codex or Cursor who need traceability, quality gates and durable project memory.
Good fit when
- An agent codes straight from a prompt and you want to steer it through a spec and tasks with coverage checks
- You want to know which decisions and bug lessons already exist and spot contradictions in old context
- A bug must follow a test-first flow and stay linked to its task
- You need one board that both a human in the browser and an agent over MCP can see
Not a fit when
- You need a light task list: 285 MCP tools eat into the agent context
- You plan to offer it to others as a hosted service: Elastic License 2.0 forbids that
- There is no Python 3.11 and no way to run Docker
Example request
Create a spec in Okto Pulse for the payments module with acceptance criteria and split it into tasks with test scenariosMCP
- Transport
- http
- Authentication
- API key
Environment variables
MCP_HOST- MCP listen address; set 0.0.0.0 for Docker
Limitations
- Python 3.11 or newer is required, and the first run downloads the `all-MiniLM-L6-v2` model. Without it the service starts, but semantic graph search is disabled.
- Elastic License 2.0 is free for personal and commercial use but forbids giving the software to third parties as a hosted service.
- The Community edition exposes all 285 tools to the agent at once, a fixed context cost described in `docs/TOKEN-USAGE.md`.
- The Codex API key is shown once and lands in shell history and Codex configuration, so keep the command private.
- The app does not clone or read source code itself: the external agent checks the environment and submits a bounded result.
How to disable. Stop `okto-pulse serve`, remove the package with `pip uninstall okto-pulse` and delete `~/.okto-pulse/`. Remove `.mcp.json` with the agent key from the project.
FAQ
How do I connect Okto Pulse to an agent?
Run `okto-pulse init` in the project: it creates `.mcp.json` with the `http://localhost:8101/mcp` address and the agent key. Claude Code picks it up when started from that folder.
Which addresses does the server use?
The web UI and API at `http://localhost:8100` and MCP at `http://localhost:8101/mcp`. Both listeners run in one Python process.
What are governance gates?
Checks on status transitions: spec coverage, task readiness, test evidence, bug traceability. There are 17 named gates, with thresholds tunable per board.
How do I run it in Docker?
The README has a `docker run` command. To expose MCP outside the container set `MCP_HOST=0.0.0.0` and publish the port.
Is the connection costly in tokens?
The connection has a fixed cost for the tool catalogue and a variable one for responses. Responses come in `summary`, `detail` and `full` projections, with measurements in `docs/TOKEN-USAGE.md`.
Discussion
No questions or reviews yet.
Related
MCP serversAn MCP server and CLI for code research: search and understand your project and all of GitHub with fewer tokens
An MCP server that turns GitHub Projects v2 into a work queue for agents: they claim tasks, report progress and spend a budget.
Tasks, docs, specs and decisions live as markdown files in your repository, and the agent reads them through the CLI and MCP
DevLens analyzes a repo once and serves agents ranked files, call flow and change impact over MCP, no file-by-file reading
Skill and plugin for Codex and Claude Code: the agent drives a project through a plan-to-maintain loop with human gates.
An MCP server that indexes a repository into a symbol graph and hands agents cards, slices and code windows under a context budget
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Okto Pulse (this card) | no | API key | 137 | |
| OpenSpec | no | 74,784 | a spec layer over the repository made of files, with no board, knowledge graph or gates on status changes | |
| Spec Kit | no | 149,633 | a set of commands for writing specs in chat, with no separate server and no 17 gates |
README in short
The README states the idea: the road from idea to code should be explicit. It then covers the pip install, agent setup, the work cycle, 17 gates, the knowledge graph, architecture, the CLI reference, Docker and data storage.