Coalent — Claude Code MCP server: what it does, how to install
Caches extracted facts for AI agents and RAG pipelines, updating entries automatically as underlying sources change.
claude mcp add --env OPENAI_API_KEY=<OPENAI_API_KEY> --env COALENT_MCP_TOKEN=<COALENT_MCP_TOKEN> --transport stdio coalent -- uvx coalentWhat Coalent does
Coalent acts as an intermediate caching layer above standard retrieval systems. Instead of storing raw document slices, it pulls atomic, verifiable statements tied to source origins. Requests match against stored meaning, which allows multiple agents to share context without running redundant completions. Multi-hop queries pull claims across distinct records without added LLM passes.
Cache units track source references. When an underlying file changes, Coalent invalidates only affected units rather than recomputing an entire knowledge structure. The v0.7 release adds a repair(read_id) routine that re-extracts omitted facts whenever an initial retrieval read comes up short.
The server expects OPENAI_API_KEY and COALENT_MCP_TOKEN in the environment. Without these values, launching the MCP runtime through uvx fails immediately.
What it helps with
- stop agents from re-reading static sources on every turn
- invalidate cached answers immediately when source files update
- gather multi-hop evidence across separate documents
- cut context token consumption during retrieval passes
Example prompts for the agent
- Check our vacation policy using the Coalent cache
- Trigger source update for confluence:hr after the recent edit
- Run repair(read_id) to re-extract facts for the failed query
How to install Coalent
Platform CLI
claude mcp add --env OPENAI_API_KEY=<OPENAI_API_KEY> --env COALENT_MCP_TOKEN=<COALENT_MCP_TOKEN> --transport stdio coalent -- uvx coalentReplace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from coalent <command>
Platform CLI
codex mcp add coalent --env OPENAI_API_KEY=<OPENAI_API_KEY> --env COALENT_MCP_TOKEN=<COALENT_MCP_TOKEN> -- uvx coalentReplace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from coalent <command>
Config file
{
"mcpServers": {
"coalent": {
"command": "uvx",
"args": [
"coalent"
],
"env": {
"OPENAI_API_KEY": "<OPENAI_API_KEY>",
"COALENT_MCP_TOKEN": "<COALENT_MCP_TOKEN>"
}
}
}
}Replace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from coalent <command>
Config file
{
"servers": {
"coalent": {
"type": "stdio",
"command": "uvx",
"args": [
"coalent"
],
"env": {
"OPENAI_API_KEY": "<OPENAI_API_KEY>",
"COALENT_MCP_TOKEN": "<COALENT_MCP_TOKEN>"
}
}
}
}Replace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from coalent <command>
Platform CLI
gemini mcp add -e OPENAI_API_KEY=<OPENAI_API_KEY> -e COALENT_MCP_TOKEN=<COALENT_MCP_TOKEN> coalent uvx coalentReplace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from coalent <command>
Step by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more
Questions about Coalent
How to install Coalent in Claude Code?
Run claude mcp add --env OPENAI_API_KEY=<OPENAI_API_KEY> --env COALENT_MCP_TOKEN=<COALENT_MCP_TOKEN> --transport stdio coalent -- uvx coalent.
What does Coalent do?
It caches claims extracted by language models according to query meaning and removes stale records whenever raw sources change.
What changed with the pool read path in v0.7?
When an embedder is configured, Coalent serves claims ranked within a token budget from a pooled store rather than a single routed entry.
